Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just... engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!show more

Daily Dose of Data Science

48,995 subscribers

23,748 Aufrufe • vor 3 Monaten •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Its not every day you wake up to find that the Pope has made your lifes work the central focus of his papacy: “Disarming AI means freeing it from the mentality of “armed” competition [..] This entails a race for ever more powerful algorithms and larger datasets, driven by the desire to secure geopolitical or commercial dominance." - POPE LEO XIV, May 2026 Here, “disarmed” means “neutralised” in the sense that this should not be a differentiator The Innovation Game (TIG) was created to keep data and algorithms open, in order to prevent monopolistic control It's not just an aspiration, it's an economic mechanism that makes open data and open algorithms the rational economic choice • All algorithms are published openly by TIG • If you are willing to make the data you process with an algorithm open, you can use it free-of-charge • Alternatively, if you would like to keep this data private, there is a fee to pay for using the algorithm • All fees are used to fund more open innovation Its an elegant, global, self-reinforcing engine The logical end point of monopoly is that innovation stops We cannot allow that to happen Pope Leo XIV I would be grateful for your thoughts on The Innovation Game

Its not every day you wake up to find that the Pope has made your lifes work the central focus of his papacy: “Disarming AI means freeing it from the mentality of “armed” competition [..] This entails a race for ever more powerful algorithms and larger datasets, driven by the desire to secure geopolitical or commercial dominance." - POPE LEO XIV, May 2026 Here, “disarmed” means “neutralised” in the sense that this should not be a differentiator The Innovation Game (TIG) was created to keep data and algorithms open, in order to prevent monopolistic control It's not just an aspiration, it's an economic mechanism that makes open data and open algorithms the rational economic choice • All algorithms are published openly by TIG • If you are willing to make the data you process with an algorithm open, you can use it free-of-charge • Alternatively, if you would like to keep this data private, there is a fee to pay for using the algorithm • All fees are used to fund more open innovation Its an elegant, global, self-reinforcing engine The logical end point of monopoly is that innovation stops We cannot allow that to happen Pope Leo XIV I would be grateful for your thoughts on The Innovation Game

John Fletcher (𝔦, 𝔦)

11,722 Aufrufe • vor 2 Monaten

Big moment for Postgres! Search has always been Postgres' weak spot, and everyone just accepted it. If you needed a real relevance-ranked keyword search, the default answer was to spin up Elasticsearch or add Algolia and deal with the data sync headaches forever. The problem isn't that Postgres can't do text search. It can. But the built-in `ts_rank` function uses a basic term frequency algorithm that doesn't come close to what modern search engines deliver. So teams end up: - Running a separate Elasticsearch cluster just for search - Building sync pipelines that inevitably drift out of consistency - Paying for managed search services that charge per query - Accepting mediocre search relevance because "good enough" ships faster But this is actually a solvable problem. You can realistically bring industry-standard search ranking directly into Postgres, which eliminates the need for external infra entirely. This exact solution is now available with the newly open-sourced pg_textsearch by Tiger Data - Creators of TimescaleDB, a Postgres extension that brings true BM25 relevance ranking into the database. BM25 is the algorithm behind Elasticsearch, Lucene, and most modern search engines. Now it runs natively in Postgres. Here's what pg_textsearch enables: - True BM25 ranking with configurable parameters (the same algorithm powering production search systems) - Simple SQL syntax: `ORDER BY content 'search terms'` - Works with Postgres text search configurations for multiple languages - Pairs naturally with pgvector for hybrid keyword + semantic search That last point matters a lot for RAG apps. The video below shows this in action, and I worked with the team to put this together. You can now do hybrid retrieval (combining keyword matching with vector similarity) in a single database, without stitching together multiple systems. The syntax is clean enough that you can add relevance-ranked search to existing queries in minutes. pg_textsearch is fully open-source under the PostgreSQL license. You can find a link to their GitHub repo in the next tweet.

Big moment for Postgres! Search has always been Postgres' weak spot, and everyone just accepted it. If you needed a real relevance-ranked keyword search, the default answer was to spin up Elasticsearch or add Algolia and deal with the data sync headaches forever. The problem isn't that Postgres can't do text search. It can. But the built-in `ts_rank` function uses a basic term frequency algorithm that doesn't come close to what modern search engines deliver. So teams end up: - Running a separate Elasticsearch cluster just for search - Building sync pipelines that inevitably drift out of consistency - Paying for managed search services that charge per query - Accepting mediocre search relevance because "good enough" ships faster But this is actually a solvable problem. You can realistically bring industry-standard search ranking directly into Postgres, which eliminates the need for external infra entirely. This exact solution is now available with the newly open-sourced pg_textsearch by Tiger Data - Creators of TimescaleDB, a Postgres extension that brings true BM25 relevance ranking into the database. BM25 is the algorithm behind Elasticsearch, Lucene, and most modern search engines. Now it runs natively in Postgres. Here's what pg_textsearch enables: - True BM25 ranking with configurable parameters (the same algorithm powering production search systems) - Simple SQL syntax: `ORDER BY content 'search terms'` - Works with Postgres text search configurations for multiple languages - Pairs naturally with pgvector for hybrid keyword + semantic search That last point matters a lot for RAG apps. The video below shows this in action, and I worked with the team to put this together. You can now do hybrid retrieval (combining keyword matching with vector similarity) in a single database, without stitching together multiple systems. The syntax is clean enough that you can add relevance-ranked search to existing queries in minutes. pg_textsearch is fully open-source under the PostgreSQL license. You can find a link to their GitHub repo in the next tweet.

Akshay 🚀

215,344 Aufrufe • vor 6 Monaten

OpenAI's Deep Research is getting a run for its money. Deep Lake was just released, and it's a different take on an AI system that can do deep research on your own data. You can use Deep Lake to build AI search with reasoning on your private and public data. (Look at the attached videos to get an idea of how it works.) If you want to research proprietary and sensitive data, Deep Research won't help you because it's limited to public data. Deep Lake, however, will allow you to use your private data. On top of that, Deep Lake supports multi-modal retrieval from the ground up. It uses vision language models for data ingestion and retrieval so that you can connect any data (PDFs, images, videos, structured data, etc.) You can even use mixed-data queries! Deep Lake can search your data from S3, Dropbox, and GCP. It learns from your queries over time, making the results as relevant to your work as possible!

OpenAI's Deep Research is getting a run for its money. Deep Lake was just released, and it's a different take on an AI system that can do deep research on your own data. You can use Deep Lake to build AI search with reasoning on your private and public data. (Look at the attached videos to get an idea of how it works.) If you want to research proprietary and sensitive data, Deep Research won't help you because it's limited to public data. Deep Lake, however, will allow you to use your private data. On top of that, Deep Lake supports multi-modal retrieval from the ground up. It uses vision language models for data ingestion and retrieval so that you can connect any data (PDFs, images, videos, structured data, etc.) You can even use mixed-data queries! Deep Lake can search your data from S3, Dropbox, and GCP. It learns from your queries over time, making the results as relevant to your work as possible!

Santiago

171,340 Aufrufe • vor 1 Jahr

forget the $699 AI pins. this $8 chip just shattered the barrier for local AI hardware. a developer just forced a 28.9 million-parameter LLM onto a standard ESP32-S3 microcontroller. it costs roughly 8 dollars, runs completely offline, and draws the power of a single LED. conventional wisdom said a model of this size simply would not fit. the chip only has 512 KB of fast SRAM and 16 MB of flash. the breakthrough is architectural. the developer moved the bulk of the embedding table into flash memory and memory-mapped it. the chip only needs to pull about 450 bytes per token, keeping the active working memory inside the fast SRAM. this means you can now embed a capable language model into a physical node for the price of two coffees. and we are already seeing the beginnings of this custom physical hardware. in the video, a creator built a minimalist voice-controlled universal remote using an ESP32. it captures voice and remotely controls the computer over bluetooth LE. he simply says "open chrome and open 20 new tabs", and the custom hardware executes it instantly. we have spent years watching model sizes explode upward. but the true frontier is the opposite direction. when an eight-dollar chip can power offline intelligence and custom physical interfaces, AI becomes local infrastructure rather than a cloud service.

forget the $699 AI pins. this $8 chip just shattered the barrier for local AI hardware. a developer just forced a 28.9 million-parameter LLM onto a standard ESP32-S3 microcontroller. it costs roughly 8 dollars, runs completely offline, and draws the power of a single LED. conventional wisdom said a model of this size simply would not fit. the chip only has 512 KB of fast SRAM and 16 MB of flash. the breakthrough is architectural. the developer moved the bulk of the embedding table into flash memory and memory-mapped it. the chip only needs to pull about 450 bytes per token, keeping the active working memory inside the fast SRAM. this means you can now embed a capable language model into a physical node for the price of two coffees. and we are already seeing the beginnings of this custom physical hardware. in the video, a creator built a minimalist voice-controlled universal remote using an ESP32. it captures voice and remotely controls the computer over bluetooth LE. he simply says "open chrome and open 20 new tabs", and the custom hardware executes it instantly. we have spent years watching model sizes explode upward. but the true frontier is the opposite direction. when an eight-dollar chip can power offline intelligence and custom physical interfaces, AI becomes local infrastructure rather than a cloud service.

ard

417,806 Aufrufe • vor 9 Tagen

The Dawn of a New Era on $SUI (9) Still in the festive spirit, let’s look at Tusky , Tusky is a storage service that's not controlled by one company. It uses something called WalrusProtocol to keep your data safe. Your data is encrypted from start to finish, so only you can see it. Instead of one place, your data goes to many different spots. This setup makes your data less likely to be lost or stolen. It helps keep your information safe and always available. Tusky gives you control over your files. You can easily manage them with the tools provided. You decide who gets to see your data. This makes it great for personal storage or working with others. It works with SuiNetwork for even more privacy. You can log in without sharing personal info, keeping everything more secure. This means only you can get to your data, with no third party involved. It is growing fast, with 100,000 uploads already. This indicates its increasing acceptance in the tech community focused on data sovereignty. It's good at managing lots of data safely. In tech, where you want to own your data, TuskyTools is popular. It gives users control over their information. This platform helps keep your data secure and gives you freedom.

The Dawn of a New Era on $SUI (9) Still in the festive spirit, let’s look at Tusky , Tusky is a storage service that's not controlled by one company. It uses something called WalrusProtocol to keep your data safe. Your data is encrypted from start to finish, so only you can see it. Instead of one place, your data goes to many different spots. This setup makes your data less likely to be lost or stolen. It helps keep your information safe and always available. Tusky gives you control over your files. You can easily manage them with the tools provided. You decide who gets to see your data. This makes it great for personal storage or working with others. It works with SuiNetwork for even more privacy. You can log in without sharing personal info, keeping everything more secure. This means only you can get to your data, with no third party involved. It is growing fast, with 100,000 uploads already. This indicates its increasing acceptance in the tech community focused on data sovereignty. It's good at managing lots of data safely. In tech, where you want to own your data, TuskyTools is popular. It gives users control over their information. This platform helps keep your data secure and gives you freedom.

Kaboom.sui🪖

19,698 Aufrufe • vor 1 Jahr

Here’s how Starlink next‑generation V3’s laser network works in space Each Starlink V3 satellite is equipped with six high-capacity space lasers capable of operating at up to 400 Gbps These laser links connect satellites directly to one another, creating a fast, high-bandwidth and resilient mesh network in orbit that can route massive amounts of data around the planet without relying entirely on nearby ground stations That means more network capacity, faster data routing, greater reliability and stronger connectivity across remote and hard-to-reach regions Starlink V3 is not just a more powerful satellite It is a major upgrade to the internet infrastructure being built in space

Here’s how Starlink next‑generation V3’s laser network works in space Each Starlink V3 satellite is equipped with six high-capacity space lasers capable of operating at up to 400 Gbps These laser links connect satellites directly to one another, creating a fast, high-bandwidth and resilient mesh network in orbit that can route massive amounts of data around the planet without relying entirely on nearby ground stations That means more network capacity, faster data routing, greater reliability and stronger connectivity across remote and hard-to-reach regions Starlink V3 is not just a more powerful satellite It is a major upgrade to the internet infrastructure being built in space

X Freeze

21,988 Aufrufe • vor 17 Tagen

Many people have asked why we no longer tag the police like before. The answer is simple: the actions of some police command is frustrating, embarrassing, disgraceful, and very discouraging. While there are many good officers, the actions of the bad ones make it hard to even appreciate the good ones. Last Saturday, Eiye cultists disrupted a sign out at Kwara Polytechnic in Ilorin, which Naija Confra reported on Sunday. They almost k!lled their rivals, but the military stepped in just in time. That attack and the current cult clash in Ilorin are being led by a man named Sanjo (the man with the black shirt and blue beret at the beginning of the video). Sanjo is the same man who k!lled people during the Ebute Metta Peace Walk in Lagos back in April, alongside another Eiye member called Baba Ikorodu (Blue j£sus). Naija Confra reported that case strongly, and it was said that the police arrested them at some point. This is not the first time Sanjo has been arrested for k!ll!ngs, but somehow he keeps getting released and returns to k!ll again. Tomorrow, he may k!ll in Ilorin, get arrested and released, then move to Lagos and k!ll again. Is this the kind of police you want us to keep tagging all the time? There are still cases where we tag them, but not always—because we’ve lost faith in some officers and their bosses. Stay safe. Reject Cultism.

Many people have asked why we no longer tag the police like before. The answer is simple: the actions of some police command is frustrating, embarrassing, disgraceful, and very discouraging. While there are many good officers, the actions of the bad ones make it hard to even appreciate the good ones. Last Saturday, Eiye cultists disrupted a sign out at Kwara Polytechnic in Ilorin, which Naija Confra reported on Sunday. They almost k!lled their rivals, but the military stepped in just in time. That attack and the current cult clash in Ilorin are being led by a man named Sanjo (the man with the black shirt and blue beret at the beginning of the video). Sanjo is the same man who k!lled people during the Ebute Metta Peace Walk in Lagos back in April, alongside another Eiye member called Baba Ikorodu (Blue j£sus). Naija Confra reported that case strongly, and it was said that the police arrested them at some point. This is not the first time Sanjo has been arrested for k!ll!ngs, but somehow he keeps getting released and returns to k!ll again. Tomorrow, he may k!ll in Ilorin, get arrested and released, then move to Lagos and k!ll again. Is this the kind of police you want us to keep tagging all the time? There are still cases where we tag them, but not always—because we’ve lost faith in some officers and their bosses. Stay safe. Reject Cultism.

Naija Confra

35,376 Aufrufe • vor 1 Jahr

The future of footwear may not be manufactured in bulk. It may be fabricated around you. That is what makes this shift so interesting to me. 3D-printed footwear is moving from novelty to a real industrial model, with market forecasts pointing to rapid growth over the next decade. At the same time, brands and manufacturers are using additive manufacturing, digital design, and custom-fit workflows to shorten development cycles and make more personalized products viable. What is new here is not just the printer. It is the system around it: → scan the foot → model the fit digitally → print the part on demand → produce closer to the customer That matters. Because once footwear becomes data-driven and locally fabricated, several things change fast: → fit gets more personal → prototyping gets faster → waste drops because you do not overproduce → inventory pressure falls because you do not need to guess demand the same way To me, that is the bigger signal. This is not just about a better sneaker. It is about a different manufacturing logic. Formlabs notes that 3D printing already enables customized orthotics with better biomechanical precision, lower material waste, and simpler digital workflows. McKinsey has also pointed to digitization and 3D design as a way to shorten design cycles and reduce sampling iterations in apparel and footwear. And once that logic matures, the use cases get much bigger: → custom athletic footwear built from gait and pressure data → hospitals producing orthotics faster and closer to the patient → micro-factories making products on demand instead of stocking shelves → footwear designed for one body, not an average body That is why I think this matters now. The question is no longer whether personalized fabrication is possible. It is whether brands move fast enough before customers start expecting every product to fit like it was made only for them. Would you actually wear a shoe fabricated around your own biometric data? #AI #3DPrinting #Footwear #Manufacturing #Innovation #FutureOfWork #RetailTech #Customization #Technology

The future of footwear may not be manufactured in bulk. It may be fabricated around you. That is what makes this shift so interesting to me. 3D-printed footwear is moving from novelty to a real industrial model, with market forecasts pointing to rapid growth over the next decade. At the same time, brands and manufacturers are using additive manufacturing, digital design, and custom-fit workflows to shorten development cycles and make more personalized products viable. What is new here is not just the printer. It is the system around it: → scan the foot → model the fit digitally → print the part on demand → produce closer to the customer That matters. Because once footwear becomes data-driven and locally fabricated, several things change fast: → fit gets more personal → prototyping gets faster → waste drops because you do not overproduce → inventory pressure falls because you do not need to guess demand the same way To me, that is the bigger signal. This is not just about a better sneaker. It is about a different manufacturing logic. Formlabs notes that 3D printing already enables customized orthotics with better biomechanical precision, lower material waste, and simpler digital workflows. McKinsey has also pointed to digitization and 3D design as a way to shorten design cycles and reduce sampling iterations in apparel and footwear. And once that logic matures, the use cases get much bigger: → custom athletic footwear built from gait and pressure data → hospitals producing orthotics faster and closer to the patient → micro-factories making products on demand instead of stocking shelves → footwear designed for one body, not an average body That is why I think this matters now. The question is no longer whether personalized fabrication is possible. It is whether brands move fast enough before customers start expecting every product to fit like it was made only for them. Would you actually wear a shoe fabricated around your own biometric data? #AI #3DPrinting #Footwear #Manufacturing #Innovation #FutureOfWork #RetailTech #Customization #Technology

Pascal Bornet

47,489 Aufrufe • vor 3 Monaten

What if you kept asking an LLM to "make it better"? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The naive approach to training for K self-improvement steps leads to K times the number of rollout steps per episode. We introduce Exploratory Iteration (ExIt), an RL-based automatic curriculum method that bootstraps diverse training distributions of self-improvement tasks by upcycling the LLM's own responses at previous turns as the starting points for both self-improvement and *self-divergence.* In order to decide what task to train on next, the curriculum prioritizes sampling of partial turn histories that led to higher return variance in its GRPO group (a learnability score that comes for free). This automatic curriculum over the bootstrapped task space teaches the model how to perform iterated self-improvement while only ever training the model on single-step self-improvement tasks. We look at ExIt's impact in both single-turn (contest math problems) and multi-turn (BFCLv3 multi-turn tasks), as well as MLE-bench, where the LLM is run in a search scaffold to produce solutions to real Kaggle competitions. Across these eval settings, we find ExIt produces models with greater capacity for inference-time self-improvement compared to GRPO. Notably, ExIt models can self-improve on test tasks for many more steps than the typical solution depth encountered during training, including a 22% improvement in MLE-bench performance compared to GRPO.

What if you kept asking an LLM to "make it better"? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The naive approach to training for K self-improvement steps leads to K times the number of rollout steps per episode. We introduce Exploratory Iteration (ExIt), an RL-based automatic curriculum method that bootstraps diverse training distributions of self-improvement tasks by upcycling the LLM's own responses at previous turns as the starting points for both self-improvement and self-divergence. In order to decide what task to train on next, the curriculum prioritizes sampling of partial turn histories that led to higher return variance in its GRPO group (a learnability score that comes for free). This automatic curriculum over the bootstrapped task space teaches the model how to perform iterated self-improvement while only ever training the model on single-step self-improvement tasks. We look at ExIt's impact in both single-turn (contest math problems) and multi-turn (BFCLv3 multi-turn tasks), as well as MLE-bench, where the LLM is run in a search scaffold to produce solutions to real Kaggle competitions. Across these eval settings, we find ExIt produces models with greater capacity for inference-time self-improvement compared to GRPO. Notably, ExIt models can self-improve on test tasks for many more steps than the typical solution depth encountered during training, including a 22% improvement in MLE-bench performance compared to GRPO.

Minqi Jiang

41,099 Aufrufe • vor 10 Monaten

A few days ago I hinted that big news is coming. Now, it’s official: I inked a partnership with @pumpdotfun to provide the most valuable trenching resources Solana has ever seen. The video below is a little taste of what’s coming. Thanks to I’ll never charge a dime for it. It’s free, and always will be. On a bittersweet note, that means I’ll no longer be active on Vector. It’s been a blast partnering with the VECTOR crew over the past few months. I’m grateful for the opportunity I had to support them and excited to see what their future holds. If you’ve been missing my broadcasts, don’t worry. I’ll save you a seat at the table. See you in the trenches again soon.

A few days ago I hinted that big news is coming. Now, it’s official: I inked a partnership with @pumpdotfun to provide the most valuable trenching resources Solana has ever seen. The video below is a little taste of what’s coming. Thanks to I’ll never charge a dime for it. It’s free, and always will be. On a bittersweet note, that means I’ll no longer be active on Vector. It’s been a blast partnering with the VECTOR crew over the past few months. I’m grateful for the opportunity I had to support them and excited to see what their future holds. If you’ve been missing my broadcasts, don’t worry. I’ll save you a seat at the table. See you in the trenches again soon.

fxnction

87,595 Aufrufe • vor 1 Jahr

Free NVIDIA GPU with 16 GB VRAM GPU for Running Local LLMs! If you want to master local LLMs but you're waiting until you can afford a $1,500 GPU, you're honestly not going to make it. The open source AI ecosystem is moving way too fast for you to wait on your budget to catch up. Especially when you can build a bleeding edge inference engine from scratch right now, completely for free. You don't need a heavy local rig to start. Google is literally letting you use an enterprise grade NVIDIA Tesla T4 GPU for $0/hour. At standard cloud computing rates (~$0.20/hr), Google Colab’s 4 hour daily free tier hands you roughly $24 worth of data center tier GPU compute every single month. And most people just waste it. Let’s talk about the hardware you get access to for free. The NVIDIA Tesla T4 is an absolute workhorse: - Architecture: NVIDIA Turing (TU104) - VRAM: 16GB GDDR6 (320 GB/s bandwidth) - Compute: 320 Tensor Cores | 2560 CUDA Cores - Performance: 130 TOPS INT8 | 8.1 TFLOPS FP32 - Power: Sipping energy at a max 70W TDP This is the exact same hardware I used to run DeepMind's Gemma 4 26B A4B QAT MoE at a 250,000 context window without a single Out Of Memory (OOM) crash. If you have a web browser and 10 minutes, you have everything you need. I’ve put together a fully documented, cell by cell Google Colab notebook that teaches you exactly how to do this. Here is what the notebook actually teaches you: - How to provision an Ubuntu Linux environment with CUDA 13.0 and verify your driver stack. - How to pull the source code and compile the latest llama.cpp C++ binaries from scratch, specifically optimizing the build for your exact GPU using the -DCMAKE_CUDA_ARCHITECTURES=native flag. - How to directly download quantized local LLMs (GGUF format) straight from HuggingFace using the CLI. - How to manage 16GB VRAM limits, offload neural network layers to the GPU, and push massive context windows. Compile raw llama.cpp, ollama run a model, or spin up the LM Studio CLI. Pick whatever stack you are comfortable with. just start building. No hardware. No credit card. No excuses. Bookmark this post right now so you don't lose the tutorial. Even if you don't have time to run it today, you are going to want this workflow in your engineering toolkit. The link to the free Colab Notebook is in the comments below. Lemme know if you need more tutorials like this.

Free NVIDIA GPU with 16 GB VRAM GPU for Running Local LLMs! If you want to master local LLMs but you're waiting until you can afford a $1,500 GPU, you're honestly not going to make it. The open source AI ecosystem is moving way too fast for you to wait on your budget to catch up. Especially when you can build a bleeding edge inference engine from scratch right now, completely for free. You don't need a heavy local rig to start. Google is literally letting you use an enterprise grade NVIDIA Tesla T4 GPU for $0/hour. At standard cloud computing rates (~$0.20/hr), Google Colab’s 4 hour daily free tier hands you roughly $24 worth of data center tier GPU compute every single month. And most people just waste it. Let’s talk about the hardware you get access to for free. The NVIDIA Tesla T4 is an absolute workhorse: - Architecture: NVIDIA Turing (TU104) - VRAM: 16GB GDDR6 (320 GB/s bandwidth) - Compute: 320 Tensor Cores | 2560 CUDA Cores - Performance: 130 TOPS INT8 | 8.1 TFLOPS FP32 - Power: Sipping energy at a max 70W TDP This is the exact same hardware I used to run DeepMind's Gemma 4 26B A4B QAT MoE at a 250,000 context window without a single Out Of Memory (OOM) crash. If you have a web browser and 10 minutes, you have everything you need. I’ve put together a fully documented, cell by cell Google Colab notebook that teaches you exactly how to do this. Here is what the notebook actually teaches you: - How to provision an Ubuntu Linux environment with CUDA 13.0 and verify your driver stack. - How to pull the source code and compile the latest llama.cpp C++ binaries from scratch, specifically optimizing the build for your exact GPU using the -DCMAKE_CUDA_ARCHITECTURES=native flag. - How to directly download quantized local LLMs (GGUF format) straight from HuggingFace using the CLI. - How to manage 16GB VRAM limits, offload neural network layers to the GPU, and push massive context windows. Compile raw llama.cpp, ollama run a model, or spin up the LM Studio CLI. Pick whatever stack you are comfortable with. just start building. No hardware. No credit card. No excuses. Bookmark this post right now so you don't lose the tutorial. Even if you don't have time to run it today, you are going to want this workflow in your engineering toolkit. The link to the free Colab Notebook is in the comments below. Lemme know if you need more tutorials like this.

Alok

178,744 Aufrufe • vor 29 Tagen

Being sore all the time is not a badge. It means: - Your volume is too high - Your recovery is in deficit - Your next session is compromised - Your nervous system is still cleaning up the last one - Your growth is being throttled by accumulated damage Soreness is a sign you did something. It is not a sign you did the right thing. A trained muscle in a sensible programme should rarely be sore at all. The first few weeks of a new stimulus, yes. After that, your body adapts. The soreness fades. The growth carries on without it. People treat this as a problem to solve. They add drop sets. They add finishers. They chase the burn. They want that next-day ache back because the ache feels like proof. It is not proof. It is damage your body now has to repair before it can build anything new. The lifters making the most progress walk out of the gym feeling worked but not wrecked. They train the same muscle again 72 hours later because they can. The sore-every-day brigade train it again in a week and call it advanced programming. It's just bad arithmetic.

Being sore all the time is not a badge. It means: - Your volume is too high - Your recovery is in deficit - Your next session is compromised - Your nervous system is still cleaning up the last one - Your growth is being throttled by accumulated damage Soreness is a sign you did something. It is not a sign you did the right thing. A trained muscle in a sensible programme should rarely be sore at all. The first few weeks of a new stimulus, yes. After that, your body adapts. The soreness fades. The growth carries on without it. People treat this as a problem to solve. They add drop sets. They add finishers. They chase the burn. They want that next-day ache back because the ache feels like proof. It is not proof. It is damage your body now has to repair before it can build anything new. The lifters making the most progress walk out of the gym feeling worked but not wrecked. They train the same muscle again 72 hours later because they can. The sore-every-day brigade train it again in a week and call it advanced programming. It's just bad arithmetic.

Sama Hoole

15,641 Aufrufe • vor 2 Monaten

He was abandoned when he needed help the most. Left alone, injured, and struggling to survive, this brave dog lost part of his body — but he never lost his will to live. Today, he is safe in rescue care, learning again what kindness feels like, wagging his tail despite everything he has been through. His journey is not defined by what he lost, but by how fiercely he keeps choosing joy. He still tries to move forward, still seeks affection, still believes in people. Now he needs our help to take the next step in his recovery — a wheelchair that will allow him to move freely, explore the world again, and live without constant strain on his body. A wheelchair means independence. It means dignity. It means this survivor can run, play, and experience life the way every dog deserves. Let’s come together and make America a better home for dogs — a place where abandoned animals are given second chances, where compassion replaces neglect, and where no survivor is left behind. If you can’t donate, please share his story. Your kindness can help change his future.

He was abandoned when he needed help the most. Left alone, injured, and struggling to survive, this brave dog lost part of his body — but he never lost his will to live. Today, he is safe in rescue care, learning again what kindness feels like, wagging his tail despite everything he has been through. His journey is not defined by what he lost, but by how fiercely he keeps choosing joy. He still tries to move forward, still seeks affection, still believes in people. Now he needs our help to take the next step in his recovery — a wheelchair that will allow him to move freely, explore the world again, and live without constant strain on his body. A wheelchair means independence. It means dignity. It means this survivor can run, play, and experience life the way every dog deserves. Let’s come together and make America a better home for dogs — a place where abandoned animals are given second chances, where compassion replaces neglect, and where no survivor is left behind. If you can’t donate, please share his story. Your kindness can help change his future.

Sofia🇵🇰

15,735 Aufrufe • vor 5 Monaten

Right now, you may not have access to models like GPT‑5.6 Sol, GPT‑4.6 Terra, GPT‑5.6 Luna, Claude Mythos 5, or Claude Fable 5. But you can run something surprisingly powerful today, locally, and completely free. in the next 10 mins on your 8 GB VRAM gaming laptop. Gemma 4 26B A4B QAT (MoE) delivers strong performance on a standard 8 GB VRAM GPU using Ollama, with no API, no usage limits, and no external dependencies. Out of the box, it reaches around 20 tokens per second without any optimizations. Only one command in your terminal: Ollama run gemma4:26b This means: Full offline capability (privacy by default) Zero recurring cost Competitive performance for many real world tasks Fast enough for interactive use on cheap consumer hardware If you're waiting for cutting edge cloud models, you're missing what is already practical today: a capable, local LLM that runs entirely on your own machine.

Right now, you may not have access to models like GPT‑5.6 Sol, GPT‑4.6 Terra, GPT‑5.6 Luna, Claude Mythos 5, or Claude Fable 5. But you can run something surprisingly powerful today, locally, and completely free. in the next 10 mins on your 8 GB VRAM gaming laptop. Gemma 4 26B A4B QAT (MoE) delivers strong performance on a standard 8 GB VRAM GPU using Ollama, with no API, no usage limits, and no external dependencies. Out of the box, it reaches around 20 tokens per second without any optimizations. Only one command in your terminal: Ollama run gemma4:26b This means: Full offline capability (privacy by default) Zero recurring cost Competitive performance for many real world tasks Fast enough for interactive use on cheap consumer hardware If you're waiting for cutting edge cloud models, you're missing what is already practical today: a capable, local LLM that runs entirely on your own machine.

Alok

65,251 Aufrufe • vor 1 Monat

New Version of HyperStore is now live! 🔥 We’re excited to announce that HyperStore has officially been upgraded to a new system version. This is not a simple UI update, it’s a full platform evolution. ⚡ What’s new? HyperStore now delivers a significantly faster and more intelligent experience powered by its rebuilt infrastructure. - 5000+ AI apps, fully structured into a living ecosystem - A new prompt-based discovery system - Faster navigation, cleaner interface, smarter results Now users don’t search for tools, they instantly reach solutions. 🧠 HyperClaw Integration HyperClaw is now fully active within HyperStore. It acts as a continuous intelligence layer that: - Keeps the platform updated in real time - Curates and optimizes AI apps dynamically - Ensures the ecosystem is always evolving 🔥 What this means? HyperStore is no longer just an AI marketplace. It is now an AI execution layer. Designed for builders, creators, and operators who move fast. 🌐 Try it now: ⚡ Find any AI solution. Instantly.

New Version of HyperStore is now live! 🔥 We’re excited to announce that HyperStore has officially been upgraded to a new system version. This is not a simple UI update, it’s a full platform evolution. ⚡ What’s new? HyperStore now delivers a significantly faster and more intelligent experience powered by its rebuilt infrastructure. - 5000+ AI apps, fully structured into a living ecosystem - A new prompt-based discovery system - Faster navigation, cleaner interface, smarter results Now users don’t search for tools, they instantly reach solutions. 🧠 HyperClaw Integration HyperClaw is now fully active within HyperStore. It acts as a continuous intelligence layer that: - Keeps the platform updated in real time - Curates and optimizes AI apps dynamically - Ensures the ecosystem is always evolving 🔥 What this means? HyperStore is no longer just an AI marketplace. It is now an AI execution layer. Designed for builders, creators, and operators who move fast. 🌐 Try it now: ⚡ Find any AI solution. Instantly.

HyperGPT

99,166 Aufrufe • vor 3 Monaten

90% of "AI developers" just download pre packaged GGUF files from Hugging Face, hit run, and call it a day. The top 10% know how to pull the raw safetensors, run the math, and quantize massive models into Q4_K_M themselves. If you think llama.cpp can only execute models, you’re missing the best part of the open source ecosystem. It’s a high performance optimization suite. Manually stripping 69% of the VRAM footprint off a brand new model architecture is where real infrastructure value is made. If you want to actually master local inference and deploy models like Google’s massive Gemma 4 12B it on consumer NVIDIA hardware using llama.cpp, you need to learn this pipeline. Let's build it. I just took the raw 22.7 GB Gemma 4 baseline and manually compressed it down to a 7.02 GB Q4_K_M GGUF artifact using llama.cpp. That is a 69% reduction in footprint. No quality loss. No VRAM bottlenecks. Just native, hardware accelerated C++ inference running a full 2,50,000 token context window on a dual NVIDIA Tesla T4 setup. Stop melting your VRAM on unoptimized weights and stop relying on other people's pipelines. Own your stack. I mapped this entire architecture from dynamic binary fetching to raw quantization and real time GPU streaming into a single, bulletproof notebook. Notebook link is in the comments below. Bookmark this blueprint for your next deployment and tell me which quantization works best for your workflow and model.

Alok

62,631 Aufrufe • vor 23 Tagen

🏝️ Bali's trash burning problem has just reached new toxic heights This week, Bali's biggest landfill has been ordered to close. The landfill is a more than 35 meter high mountain of decaying trash covering 32 hectares. It's closed because the gases emanated from it are toxic and ground water is being contaminated This means there's no other landfill of this size near to accept Bali's current trash. So the government has instructed people that "household waste should be disposed of at home" which in Bali means burning it! Trash burning has been a tradition for centuries in Bali (and Indonesia), but it used be mostly organic matter that was being burned Once plastic arrived the tradition didn't change though and locals started burning plastic too, pumping toxic gases into the air causing massive spikes of lung cancer and other respiratory diseases The tradition means on a daily basis around 7 in the morning and 6 in the evening, your neighbours will be burning their trash, and if you're not lucky there's a construction site near which will burn even more but all day! Burning plastic means you emit dioxins and furans, some of the most potent human carcinogens that exist I have friends in Indonesia with family with lung cancer cases, it's a real thing. My fear with Bali has always been that the digital nomads there might not realize the slow danger creeping up on them. You get used to the trash burning and polluted air very quickly in Bali and it probably doesn't affect you over a year or so But if you're there for many years, it will in some way or the other! For a community that's so about fitness and health, air quality is weirdly a consideration mostly overlooked in Bali

🏝️ Bali's trash burning problem has just reached new toxic heights This week, Bali's biggest landfill has been ordered to close. The landfill is a more than 35 meter high mountain of decaying trash covering 32 hectares. It's closed because the gases emanated from it are toxic and ground water is being contaminated This means there's no other landfill of this size near to accept Bali's current trash. So the government has instructed people that "household waste should be disposed of at home" which in Bali means burning it! Trash burning has been a tradition for centuries in Bali (and Indonesia), but it used be mostly organic matter that was being burned Once plastic arrived the tradition didn't change though and locals started burning plastic too, pumping toxic gases into the air causing massive spikes of lung cancer and other respiratory diseases The tradition means on a daily basis around 7 in the morning and 6 in the evening, your neighbours will be burning their trash, and if you're not lucky there's a construction site near which will burn even more but all day! Burning plastic means you emit dioxins and furans, some of the most potent human carcinogens that exist I have friends in Indonesia with family with lung cancer cases, it's a real thing. My fear with Bali has always been that the digital nomads there might not realize the slow danger creeping up on them. You get used to the trash burning and polluted air very quickly in Bali and it probably doesn't affect you over a year or so But if you're there for many years, it will in some way or the other! For a community that's so about fitness and health, air quality is weirdly a consideration mostly overlooked in Bali

@levelsio

240,616 Aufrufe • vor 11 Monaten

A few points on the Powering Canada Strong announcement that is important to understand; * Doubling Canada's electricity generation capacity is paramount. I just wish it wouldn't take 20+ years. We don't generate enough electricity to be self-sufficient or participate in future industries. We have no choice. Has to be done. It's something I called for a while and spoke on. * Linking the connectivity of Canada's fragmented grid. This is a must to increase productivity, and remove waste. It's a one step back for two steps forward type of investment. * the connection and expansion of the grid is one of the important things we need to do reach mining areas and develop these sectors and for the growth of smaller communities around. The problem with these whole announcement is that it is all net zero based which means it won't necessarily build the most reliable possible grid for the $ and will other ridiculous costs to be carbon tax trading based on the way. It's completely inefficient from capital planning point. Mark Carney says: It will require the spreading of costs over time using our AAA balance sheet so that ratepayers don't pay all of the costs of investments today. That means the government is planning to borrow MASSIVELY! That cost will appear not only in your electricity bill but also in the value of the CAD and interest costs that is already hitting record every single year. This plan is utilizing legitimate needed action to transform all of Canada's energy need into ideological driven carbon tax trade system and inefficient power generation that all together will cost Canadian taxpayers hundreds of billions more than it should.

A few points on the Powering Canada Strong announcement that is important to understand; * Doubling Canada's electricity generation capacity is paramount. I just wish it wouldn't take 20+ years. We don't generate enough electricity to be self-sufficient or participate in future industries. We have no choice. Has to be done. It's something I called for a while and spoke on. * Linking the connectivity of Canada's fragmented grid. This is a must to increase productivity, and remove waste. It's a one step back for two steps forward type of investment. * the connection and expansion of the grid is one of the important things we need to do reach mining areas and develop these sectors and for the growth of smaller communities around. The problem with these whole announcement is that it is all net zero based which means it won't necessarily build the most reliable possible grid for the $ and will other ridiculous costs to be carbon tax trading based on the way. It's completely inefficient from capital planning point. Mark Carney says: It will require the spreading of costs over time using our AAA balance sheet so that ratepayers don't pay all of the costs of investments today. That means the government is planning to borrow MASSIVELY! That cost will appear not only in your electricity bill but also in the value of the CAD and interest costs that is already hitting record every single year. This plan is utilizing legitimate needed action to transform all of Canada's energy need into ideological driven carbon tax trade system and inefficient power generation that all together will cost Canadian taxpayers hundreds of billions more than it should.

Kirk Lubimov

24,482 Aufrufe • vor 2 Monaten

Dawah is deception. Why are you leaving out the context SeerahPro ? You know right after this he says he doesn’t remember well what he wrote in that entry on Muhaymin. Why leave that part out? He even goes on to say that section seems to be non-committal and he needs to go back and re-read it. So since you don’t want to show every one the full context, I’ll do it, and expose how you are lying now and lied in the debate. Sinai doesn’t help your claim that the Quran believes the previous scriptures were textually corrupted. He says he doesn’t think the Quran believes the previous scriptures were corrupted in the deep past and that they are unrecoverable (Clip 1). Now let’s look at the short clip you pulled in context (Clip 2). First, what he says does not affirm why you cited him to begin with, which is that the Quran is supposed to be a supreme authority and determine which parts of our scriptures are corrupt and which are not. In the very clip you pulled he says, “The Quran is claiming to be able to sort of determine the correct reading of these previous scriptures and to remedy misreading and misunderstandings.” So by arbiter he means it is not determining what parts of our scriptures are true or corrupted, it is only an arbiter in determining meaning. You just buried yourself by citing this, because you implied Sinai was saying the Quran is an arbiter for determining textual corruption, which he doesn’t say. It’s like you didn’t even pay attention to what you pulled. That is embarrassing. Second, right after this (in what you left out) the interviewer reminds him he entertained two possibilities in his book and was noncommittal. Sinai says he doesn’t remember what he said and was probably leaving open possibilities, which is what I already had to correct you on. If you go back to the book, once again, he said on page 707 essentially what I said, “Muhaymin, meanwhile, is derived from Syriac mhaymnā (or conceivably from its equivalent in some other form of Aramaic), a passive participle of the verb haymen and meaning ‘trustworthy, faithful, loyal.’” After this all he does in the book is mention other possibilities and the one you want is more contingent on non-canonical readings. But even if that interpretation is correct (which Sinai confirms in this interview is his view) it doesn’t help you, because you misunderstood what Sinai means by ‘arbiter.’ In Clip 3 he first says, “So I need to reread that entry. I guess it seems to be very non-committal.” But remember you said (Clip 4), “In his 'Key Terms of the Quran' he explicitly says that Muhaymin means the Quran is the arbiter of the contents.” That is incorrect. He did not explicitly say that. Why did you lie? In Clip 3 he goes on to say “But I think entrusted with authority can mean, authorized to sort of authoritatively interpret and determine especially theological significance.” So you don’t even understand what Sinai means by “arbiter.” He doesn’t mean what you mean, that the Quran is determining which parts of our scriptures are true and corrupted, but that it is an arbiter of meaning. The fact that you still do not see this means you are beyond delusional. You never should have cited Sinai because he doesn’t help you at all. The funny thing is we were already planning to go live tomorrow to review this interview because he doesn’t help you at all and unintentionally helps with the Islamic Dilemma.

Dawah is deception. Why are you leaving out the context SeerahPro ? You know right after this he says he doesn’t remember well what he wrote in that entry on Muhaymin. Why leave that part out? He even goes on to say that section seems to be non-committal and he needs to go back and re-read it. So since you don’t want to show every one the full context, I’ll do it, and expose how you are lying now and lied in the debate. Sinai doesn’t help your claim that the Quran believes the previous scriptures were textually corrupted. He says he doesn’t think the Quran believes the previous scriptures were corrupted in the deep past and that they are unrecoverable (Clip 1). Now let’s look at the short clip you pulled in context (Clip 2). First, what he says does not affirm why you cited him to begin with, which is that the Quran is supposed to be a supreme authority and determine which parts of our scriptures are corrupt and which are not. In the very clip you pulled he says, “The Quran is claiming to be able to sort of determine the correct reading of these previous scriptures and to remedy misreading and misunderstandings.” So by arbiter he means it is not determining what parts of our scriptures are true or corrupted, it is only an arbiter in determining meaning. You just buried yourself by citing this, because you implied Sinai was saying the Quran is an arbiter for determining textual corruption, which he doesn’t say. It’s like you didn’t even pay attention to what you pulled. That is embarrassing. Second, right after this (in what you left out) the interviewer reminds him he entertained two possibilities in his book and was noncommittal. Sinai says he doesn’t remember what he said and was probably leaving open possibilities, which is what I already had to correct you on. If you go back to the book, once again, he said on page 707 essentially what I said, “Muhaymin, meanwhile, is derived from Syriac mhaymnā (or conceivably from its equivalent in some other form of Aramaic), a passive participle of the verb haymen and meaning ‘trustworthy, faithful, loyal.’” After this all he does in the book is mention other possibilities and the one you want is more contingent on non-canonical readings. But even if that interpretation is correct (which Sinai confirms in this interview is his view) it doesn’t help you, because you misunderstood what Sinai means by ‘arbiter.’ In Clip 3 he first says, “So I need to reread that entry. I guess it seems to be very non-committal.” But remember you said (Clip 4), “In his 'Key Terms of the Quran' he explicitly says that Muhaymin means the Quran is the arbiter of the contents.” That is incorrect. He did not explicitly say that. Why did you lie? In Clip 3 he goes on to say “But I think entrusted with authority can mean, authorized to sort of authoritatively interpret and determine especially theological significance.” So you don’t even understand what Sinai means by “arbiter.” He doesn’t mean what you mean, that the Quran is determining which parts of our scriptures are true and corrupted, but that it is an arbiter of meaning. The fact that you still do not see this means you are beyond delusional. You never should have cited Sinai because he doesn’t help you at all. The funny thing is we were already planning to go live tomorrow to review this interview because he doesn’t help you at all and unintentionally helps with the Islamic Dilemma.

InspiringPhilosophy - Michael Jones

36,857 Aufrufe • vor 19 Tagen