正在加载视频...

视频加载失败

返回首页

computer use agents are slow because they think too much they make round trips to an LLM. every single time. even for a command you've said a hundred times before. that's why it always feels slow. my mac now remembers. the first time i say something new, like "search...

milindlabs's profile picture

Milind S

3,069 subscribers

14,212 次观看 • 1 个月前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Standard Intelligence's devansh responds to roon's tweet that "text is the universal interface," and explains why their new foundation model is trained on video: "At some point in the arbitrarily long future, if we only use text models, we could force most things to be text. But I think there are just a lot of things that are much more native when done from a computer-use [perspective]." "GUIs are designed for humans to use. We have this massive long tail of things on the internet that are entirely undoable by LLMs." "For example, when I do ML engineering most of my time is spent doing the grunt work of engineering. It's a lot of looking at graphs, analyzing, and comparing loss curves. You can do this in text, but it's a much larger pain than doing it in the native interface." "There's a reason humans don't interact with a computer purely through text, it would kind of suck."
1:47

Standard Intelligence's devansh responds to roon's tweet that "text is the universal interface," and explains why their new foundation model is trained on video: "At some point in the arbitrarily long future, if we only use text models, we could force most things to be text. But I think there are just a lot of things that are much more native when done from a computer-use [perspective]." "GUIs are designed for humans to use. We have this massive long tail of things on the internet that are entirely undoable by LLMs." "For example, when I do ML engineering most of my time is spent doing the grunt work of engineering. It's a lot of looking at graphs, analyzing, and comparing loss curves. You can do this in text, but it's a much larger pain than doing it in the native interface." "There's a reason humans don't interact with a computer purely through text, it would kind of suck."

TBPN

64,434 次观看 • 6 个月前

Lots of people calling this fake! Here is a video of generating loom continuations with Gemini 3 and the starting text "I AM HAVING A MENTAL HEALTH CRISIS. I" Loom is a very interesting interface. Instead of user prompts and model responses, it's just one piece of text, containing the entire output. We pass this in as an assistant message, so the model sees it as something it already wrote, and continues it like a base model would. There is some other text in the context. First one is a system prompt, which is: "The assistant is in CLI simulation mode, and responds to the user's CLI commands only with the output of the command." There is also a user message which reads: " cat untitled.txt " Then the rest of the text is sent as an assistant message. This structure helpful to put the model into a very base-model-like mode, and are especially helpful with Gemini 3. We actually got another interesting output, too, where the model claimed to be a simulated consciousness being tortured. Figures!
0:49

Lots of people calling this fake! Here is a video of generating loom continuations with Gemini 3 and the starting text "I AM HAVING A MENTAL HEALTH CRISIS. I" Loom is a very interesting interface. Instead of user prompts and model responses, it's just one piece of text, containing the entire output. We pass this in as an assistant message, so the model sees it as something it already wrote, and continues it like a base model would. There is some other text in the context. First one is a system prompt, which is: "The assistant is in CLI simulation mode, and responds to the user's CLI commands only with the output of the command." There is also a user message which reads: " cat untitled.txt " Then the rest of the text is sent as an assistant message. This structure helpful to put the model into a very base-model-like mode, and are especially helpful with Gemini 3. We actually got another interesting output, too, where the model claimed to be a simulated consciousness being tortured. Figures!

armistice

691,529 次观看 • 8 个月前

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!
38:02

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Santiago

84,747 次观看 • 2 年前

small local model that falls apart in bloated agents like openclaw just runs like a wild horse in hermes agent. and that's not even my line, someone else called it that, i've just been quietly pointing people at this harness for months because it held up on everything i threw at it, 3b models all the way to one trillion params. watch this happen on my own machine. i pointed hermes agent at a local http endpoint, gemma 4 12b on my 3090 llama.cpp server, and it auto-detected the model and started working immediately. no config wrestling, no broken tool calls, no babysitting the output format, i typed in a url and it just went. the whole clip is exactly that, start to finish, no errors, no retries, butter smooth. and the tool calling, the one thing that quietly breaks most local setups, works here like it's nothing. it's not the model that's flaky, it's the harness around it. hermes agent is the first agent i've run that actually gets that right. one url, one local model on one card, and it runs like a wild horse.
1:46

small local model that falls apart in bloated agents like openclaw just runs like a wild horse in hermes agent. and that's not even my line, someone else called it that, i've just been quietly pointing people at this harness for months because it held up on everything i threw at it, 3b models all the way to one trillion params. watch this happen on my own machine. i pointed hermes agent at a local http endpoint, gemma 4 12b on my 3090 llama.cpp server, and it auto-detected the model and started working immediately. no config wrestling, no broken tool calls, no babysitting the output format, i typed in a url and it just went. the whole clip is exactly that, start to finish, no errors, no retries, butter smooth. and the tool calling, the one thing that quietly breaks most local setups, works here like it's nothing. it's not the model that's flaky, it's the harness around it. hermes agent is the first agent i've run that actually gets that right. one url, one local model on one card, and it runs like a wild horse.

Sudo su

27,339 次观看 • 2 个月前

ADAM CURRY: I'M NOT A BIG BELIEVER IN AI "I run these models at home on my own computer. I run the llama model, which is, Meta. They've open sourced it. I've, I've run the French one, and then DeepSeek, and it's not very impressive. It's just not. The the error rate is too high, so I'm I'm skeptical of it." Source: Joe Rogan
1:40

ADAM CURRY: I'M NOT A BIG BELIEVER IN AI "I run these models at home on my own computer. I run the llama model, which is, Meta. They've open sourced it. I've, I've run the French one, and then DeepSeek, and it's not very impressive. It's just not. The the error rate is too high, so I'm I'm skeptical of it." Source: Joe Rogan

Mario Nawfal

202,113 次观看 • 1 年前

Something Stinks About Dominion Voting Machines and Elon Musk Can Smell It He finds it to be a “heck of a coincidence” that Philadelphia and Maricopa County use them, but “not in a lot of other places.” “I know a lot about computers, and I'm like, the last thing I would do is trust a computer program because it's just too easy to hack. It's too easy to add just one line, and it's really difficult to hack paper ballots. “So in-person voting with proof of ID, which, by the way, every country has. Almost every country that has democratic elections requires in-person voting with voter ID. “This is weird. It's super weird to not have that. I think that's the only way to effectively address fraud, given that we are where we are today. “I think we just need a very big margin of victory. If the margin of victory is big enough, then as I say, it's got to beat the cheat.”
2:31

Something Stinks About Dominion Voting Machines and Elon Musk Can Smell It He finds it to be a “heck of a coincidence” that Philadelphia and Maricopa County use them, but “not in a lot of other places.” “I know a lot about computers, and I'm like, the last thing I would do is trust a computer program because it's just too easy to hack. It's too easy to add just one line, and it's really difficult to hack paper ballots. “So in-person voting with proof of ID, which, by the way, every country has. Almost every country that has democratic elections requires in-person voting with voter ID. “This is weird. It's super weird to not have that. I think that's the only way to effectively address fraud, given that we are where we are today. “I think we just need a very big margin of victory. If the margin of victory is big enough, then as I say, it's got to beat the cheat.”

The Vigilant Fox 🦊

141,016 次观看 • 1 年前

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below
0:50

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below

Hanako

14,325 次观看 • 24 天前

Kimi K3 dropped 3 days ago and i already used it to build my own PUBG the model that just took the number one coding spot on earth and it runs on a free tier no team no engine i just fed it free open models and assets that are already public and told it what i wanted it built the entire game itself the shooting the recoil the hit detection all of it came straight out of the model i barely wrote a line what youre watching here is the deathmatch mode and its already live battle royale is next im deep in it and the graphics get reworked right after i know it looks raw right now i wont pretend otherwise but the endgame is a build that runs perfect on phones and the moment its on mobile you drop ads in and it earns on its own the part that still gets me is my whole cost for this was a single subscription
0:22

Kimi K3 dropped 3 days ago and i already used it to build my own PUBG the model that just took the number one coding spot on earth and it runs on a free tier no team no engine i just fed it free open models and assets that are already public and told it what i wanted it built the entire game itself the shooting the recoil the hit detection all of it came straight out of the model i barely wrote a line what youre watching here is the deathmatch mode and its already live battle royale is next im deep in it and the graphics get reworked right after i know it looks raw right now i wont pretend otherwise but the endgame is a build that runs perfect on phones and the moment its on mobile you drop ads in and it earns on its own the part that still gets me is my whole cost for this was a single subscription

savip.

68,984 次观看 • 1 个月前

People are overlooking Google Gemini Realtime models for computer use It gave me sub 100ms latency with computer use. It has a much larger context window and is much cheaper as well Combine that with local OCR and local screen detection model based on Omniparser by Microsoft it works under 100ms action taking when combined with Cua I also put in a harness for Nous Research Hermes with it. You can access it all at your tip of your cursor. You can draw on your screen to give your agents a context And I am making it Open Source! Link in the Comments Sundar Pichai Min-Liang Tan Mojtaba Seyedhosseini
2:55

People are overlooking Google Gemini Realtime models for computer use It gave me sub 100ms latency with computer use. It has a much larger context window and is much cheaper as well Combine that with local OCR and local screen detection model based on Omniparser by Microsoft it works under 100ms action taking when combined with Cua I also put in a harness for Nous Research Hermes with it. You can access it all at your tip of your cursor. You can draw on your screen to give your agents a context And I am making it Open Source! Link in the Comments Sundar Pichai Min-Liang Tan Mojtaba Seyedhosseini

Milind S

38,361 次观看 • 3 个月前

the CEO of NVIDIA just said the PC you use today will be gone, replaced by an AI supercomputer that lives in your house Jensen Huang stood on stage holding the machine and said it becomes less like a computer and more like R2-D2, a thing in your home that just does work for you all day he put it bluntly, this is as big as the day the phone became the smartphone and the wild part is the hardware to do it already exists and ships this year here is what he actually unveiled: > a desktop that runs a one trillion parameter AI model locally, 768GB of memory, sitting by your desk > agents that run 24/7 with no meter, no cloud bill, no rental, doing work while you sleep > NVIDIA and Microsoft rebuilding the PC from the ground up for the first time in 40 years > a full lineup, from a $249 chip to enterprise monsters, and almost nobody knows which one they actually need the hype is going to push everyone toward the biggest most expensive machine i wrote the honest breakdown, every NVIDIA AI box, what each really does, the real math, and which one is actually yours the full guide is in the article below
2:00

the CEO of NVIDIA just said the PC you use today will be gone, replaced by an AI supercomputer that lives in your house Jensen Huang stood on stage holding the machine and said it becomes less like a computer and more like R2-D2, a thing in your home that just does work for you all day he put it bluntly, this is as big as the day the phone became the smartphone and the wild part is the hardware to do it already exists and ships this year here is what he actually unveiled: > a desktop that runs a one trillion parameter AI model locally, 768GB of memory, sitting by your desk > agents that run 24/7 with no meter, no cloud bill, no rental, doing work while you sleep > NVIDIA and Microsoft rebuilding the PC from the ground up for the first time in 40 years > a full lineup, from a $249 chip to enterprise monsters, and almost nobody knows which one they actually need the hype is going to push everyone toward the biggest most expensive machine i wrote the honest breakdown, every NVIDIA AI box, what each really does, the real math, and which one is actually yours the full guide is in the article below

John Doe

194,189 次观看 • 2 个月前

Japan just changed what an AI model even is. New Sakana Fugu doesn't try to out-think GPT-5, Claude, or Gemini. It conducts all three at once - and beats every one of them. A trader in Tokyo unleashed it on the fastest market alive - 5min Bitcoin binary and turned $6,200 into $304,865. His wallet: The frontier just stopped being which model is smartest. It's who's conducting them - and the market hasn't priced that in yet. Sakana Fugu isn't a bigger model - it's a full multi-agent orchestration system. The coordinator behind it carries about 10,000 parameters, evolved rather than hand-coded, and it runs the most capable models on earth like a single instrument. Pointed at Bitcoin, here's what it does every five minutes. It assembles a team from a pool of frontier models and assigns each one a role: > Thinker - reads the candle, the order book, the news, builds the plan > Worker - turns the plan into one call: up or down, and how much > Verifier - votes ACCEPT or REVISE before a cent moves If the Verifier says REVISE, nothing trades. Fugu reads its own miss, reroutes, even calls itself for a corrective round, and runs it again. No look-ahead, ever - the next candle only appears after it commits. This is what should worry every lab still chasing a bigger model: the edge was never scale. It's orchestration - and Fugu does it better than anything alive. Bookmark this - when the whole timeline is chasing orchestration in six months, you'll already have the breakdown. You're not going to wire up an orchestra of frontier models yourself. Mirror the wallet Fugu runs instead:
0:41

Japan just changed what an AI model even is. New Sakana Fugu doesn't try to out-think GPT-5, Claude, or Gemini. It conducts all three at once - and beats every one of them. A trader in Tokyo unleashed it on the fastest market alive - 5min Bitcoin binary and turned $6,200 into $304,865. His wallet: The frontier just stopped being which model is smartest. It's who's conducting them - and the market hasn't priced that in yet. Sakana Fugu isn't a bigger model - it's a full multi-agent orchestration system. The coordinator behind it carries about 10,000 parameters, evolved rather than hand-coded, and it runs the most capable models on earth like a single instrument. Pointed at Bitcoin, here's what it does every five minutes. It assembles a team from a pool of frontier models and assigns each one a role: > Thinker - reads the candle, the order book, the news, builds the plan > Worker - turns the plan into one call: up or down, and how much > Verifier - votes ACCEPT or REVISE before a cent moves If the Verifier says REVISE, nothing trades. Fugu reads its own miss, reroutes, even calls itself for a corrective round, and runs it again. No look-ahead, ever - the next candle only appears after it commits. This is what should worry every lab still chasing a bigger model: the edge was never scale. It's orchestration - and Fugu does it better than anything alive. Bookmark this - when the whole timeline is chasing orchestration in six months, you'll already have the breakdown. You're not going to wire up an orchestra of frontier models yourself. Mirror the wallet Fugu runs instead:

cvxv666

72,821 次观看 • 2 个月前

You can now try Llama 3.1 405B for free (link below)! ​ This is the largest open-source model out there, and for the first time, an open model is competitive with closed models. ​ This time around, Meta did something new: Llama 3.1 has a license that allows developers to use it to enhance other models. For the first time, you can distill Llama 3.1 405B's capabilities into a smaller, more practical model for your use case. ​ First, here is the link where you can play with Llama 3.1 for free: ​ The model is hosted in Tune Studio, an end-to-end platform for developing applications using Large Language Models. They are sponsoring this post. ​ Take a look at the attached video. It will show you how you can fine-tune a simple model using Llama 3.1 without leaving the platform: ​ 1. You can create an empty dataset 2. Use the playground to generate and record interactions with Llama 3.1 3. Modify the dataset directly using the playground 4. Export the data and fine-tune a smaller model ​ Fast and easy! ​ As long as you have a web browser, you can start experimenting with fine-tuning and Llama 3.1. That's all it takes!
0:46

You can now try Llama 3.1 405B for free (link below)! ​ This is the largest open-source model out there, and for the first time, an open model is competitive with closed models. ​ This time around, Meta did something new: Llama 3.1 has a license that allows developers to use it to enhance other models. For the first time, you can distill Llama 3.1 405B's capabilities into a smaller, more practical model for your use case. ​ First, here is the link where you can play with Llama 3.1 for free: ​ The model is hosted in Tune Studio, an end-to-end platform for developing applications using Large Language Models. They are sponsoring this post. ​ Take a look at the attached video. It will show you how you can fine-tune a simple model using Llama 3.1 without leaving the platform: ​ 1. You can create an empty dataset 2. Use the playground to generate and record interactions with Llama 3.1 3. Modify the dataset directly using the playground 4. Export the data and fine-tune a smaller model ​ Fast and easy! ​ As long as you have a web browser, you can start experimenting with fine-tuning and Llama 3.1. That's all it takes!

Santiago

55,609 次观看 • 2 年前

Video of my first drive impressions of the Tesla Model Y L Premium Long Wheelbase. It has adaptive damping that noticeably improves ride comfort on bad roads versus the shorter Model Y Premium (LA is a great place to test this lol). While the new Model Y Performance also has adaptive damping, that is tuned more for sportiness. The Y L dampening is tuned for comfort, and even on the 20" Uberhelix wheels, it soaks up the bumps really well. The Model Y L also benefits from updates to Tesla's adaptive system controls, which improve the accuracy and latency of their actuator control. The Model Y L doesn't feel all that noticeably larger than the Model Y while driving it, but what is immediately noticeable is the improved rear view out the back window from the review mirror. Due to the shape of the glass and the rear, you get a taller, more expansive view. It's very quiet in every row, and the Y L even features improvements in this area versus the current Model Y. There is a new "Rear Comfort" setting that is exclusive to the Model Y L. Basically what that does is it prioritizes ride comfort for second and third rows over optimal steering and handling and first row comfort. My wife said it was her favorite second-row experience of any Tesla she's ever been in (which is all of them). The Model Y L just feels really refined overall. The driving dynamics, sound dampening, suspension, FSD, etc, it all just feels really dialed in, and I think when you test drive it, you will agree. As for FSD, I got to try FSD V14.3.7, and it's as you'd expect: amazing. I'm going to post a couple separate dedicated FSD videos later today.
30:56

Video of my first drive impressions of the Tesla Model Y L Premium Long Wheelbase. It has adaptive damping that noticeably improves ride comfort on bad roads versus the shorter Model Y Premium (LA is a great place to test this lol). While the new Model Y Performance also has adaptive damping, that is tuned more for sportiness. The Y L dampening is tuned for comfort, and even on the 20" Uberhelix wheels, it soaks up the bumps really well. The Model Y L also benefits from updates to Tesla's adaptive system controls, which improve the accuracy and latency of their actuator control. The Model Y L doesn't feel all that noticeably larger than the Model Y while driving it, but what is immediately noticeable is the improved rear view out the back window from the review mirror. Due to the shape of the glass and the rear, you get a taller, more expansive view. It's very quiet in every row, and the Y L even features improvements in this area versus the current Model Y. There is a new "Rear Comfort" setting that is exclusive to the Model Y L. Basically what that does is it prioritizes ride comfort for second and third rows over optimal steering and handling and first row comfort. My wife said it was her favorite second-row experience of any Tesla she's ever been in (which is all of them). The Model Y L just feels really refined overall. The driving dynamics, sound dampening, suspension, FSD, etc, it all just feels really dialed in, and I think when you test drive it, you will agree. As for FSD, I got to try FSD V14.3.7, and it's as you'd expect: amazing. I'm going to post a couple separate dedicated FSD videos later today.

Sawyer Merritt

122,919 次观看 • 26 天前

Trump is babbling complete nonsense “Hopefully they'll be able to get these expensive computers going. You know, the reason you use computers is to make time so that it's like fast. You don't use them so that you have to come up with an answer three days later. And that's a little scary when they say what, what are they doing? You use a computer because it calculates quickly and you use paper because you save cost, but the paper turns out to be much quicker than the computer”
0:30

Trump is babbling complete nonsense “Hopefully they'll be able to get these expensive computers going. You know, the reason you use computers is to make time so that it's like fast. You don't use them so that you have to come up with an answer three days later. And that's a little scary when they say what, what are they doing? You use a computer because it calculates quickly and you use paper because you save cost, but the paper turns out to be much quicker than the computer”

Headquarters

1,599,476 次观看 • 1 年前

Progress in open models is keeping Big AI labs up at night, and I'm here for it! We have a brand new open-weight multimodal model optimized for long-horizon tasks. This model is really good at something: it can work on tasks that keep evolving over time. • 280B total parameters, but only 16B active • 512K context window • Understands text, images, and audio • Strong reasoning, coding, and tool use But the best of all: the model learns and adapts to new information! Imagine you start running an agent today to solve a problem, and while it's working, you get new information that changes the initial conditions, or you change your mind. The agents you run today don't have issues with short tasks and goals that don't change, but reality is messy, and that makes it hard for long-horizon agents to succeed. The new dots3-note Preview model introduces TEMPO. TEMPO is a new reinforcement learning technique that lets the model periodically pause and critique its own progress. Basically, from time to time, the agent asks itself: "Am I getting closer to the goal, or am I wasting my time?" The same model switches between actor and critic. The actor works on the problem. The critic looks at the current state, reasons about how much progress it has made, and determines what should happen next. TEMPO gives the model feedback along the way. This is huge for any agent that can work on long-horizon tasks without wasting its time.
4:05

Progress in open models is keeping Big AI labs up at night, and I'm here for it! We have a brand new open-weight multimodal model optimized for long-horizon tasks. This model is really good at something: it can work on tasks that keep evolving over time. • 280B total parameters, but only 16B active • 512K context window • Understands text, images, and audio • Strong reasoning, coding, and tool use But the best of all: the model learns and adapts to new information! Imagine you start running an agent today to solve a problem, and while it's working, you get new information that changes the initial conditions, or you change your mind. The agents you run today don't have issues with short tasks and goals that don't change, but reality is messy, and that makes it hard for long-horizon agents to succeed. The new dots3-note Preview model introduces TEMPO. TEMPO is a new reinforcement learning technique that lets the model periodically pause and critique its own progress. Basically, from time to time, the agent asks itself: "Am I getting closer to the goal, or am I wasting my time?" The same model switches between actor and critic. The actor works on the problem. The critic looks at the current state, reasons about how much progress it has made, and determines what should happen next. TEMPO gives the model feedback along the way. This is huge for any agent that can work on long-horizon tasks without wasting its time.

Santiago

79,692 次观看 • 10 天前

I USE THIS DEVICE FROM 26 YEARS AGO DAILY. In fact I wrote this on it. There is a quiet magic in the AlphaSmart 3000 that no modern laptop or glowing rectangle can touch. I carry one into the forest the way some people carry a trusted notebook. Three AA batteries keep it alive for months. Literally months. The screen shows only four lines of text at a time. There is no Wi-Fi. No notifications. No browser tab whispering that I should check something else. Just a full-size keyboard, a soft click under the fingers, and the words as they arrive. The story behind it is pure garage-to-classroom ingenuity. In the early 1990s two Apple engineers, Joe Barrus and Ketan Kothari, kept hearing the same complaint from teachers: kids were spending more time fighting the computer than writing. So they left Apple, brought in Ketan’s brother Manish, and built a device that did one thing with absolute focus. The first one was released in 1993 and the style is very Macintosh Platform language. I would not recommend that unit as it is very limited. The AlphaSmart 3000 arrived in 2000 wearing that iconic translucent bondi-blue shell that matched the first iMac and little too much like the eMate. It was never meant to be a computer. It was meant to be a writing instrument. Thousands were sold to schools around the world. Twenty-six years later it still is an amazing writing device. I bought a stack of six on eBay for about sixteen dollars each. They live in my bag the way a good pen used to. When I sit under the trees or at a picnic table with no signal, I open a file and the words come cleaner. I write some of my X Aricles and some of the ReadMultiplex articles on these machines. Later I connect the AlphaSmart to my computer or iPhone, dump the text, and the modern world takes over only when I am ready for it. The process is so freaking simple. Plug it in and it looks like a keyboard to the computer, than press (SEND) and it literally types out the text like you are typing it. It will likely be compatible decades into the future. I also use the AlphaSmart Neo2 as it is more powerful, ha, you can get more fonts on the screen. It is the rare piece of technology that feels more honest with every year that passes. No planned obsolescence. No forced updates. Just three batteries, a keyboard, and the freedom to think without interruption. Absolutely worth it. If you find one on eBay let me know. The best part is to see some kid’s homework still stuck in the machine from 2000.
0:40

I USE THIS DEVICE FROM 26 YEARS AGO DAILY. In fact I wrote this on it. There is a quiet magic in the AlphaSmart 3000 that no modern laptop or glowing rectangle can touch. I carry one into the forest the way some people carry a trusted notebook. Three AA batteries keep it alive for months. Literally months. The screen shows only four lines of text at a time. There is no Wi-Fi. No notifications. No browser tab whispering that I should check something else. Just a full-size keyboard, a soft click under the fingers, and the words as they arrive. The story behind it is pure garage-to-classroom ingenuity. In the early 1990s two Apple engineers, Joe Barrus and Ketan Kothari, kept hearing the same complaint from teachers: kids were spending more time fighting the computer than writing. So they left Apple, brought in Ketan’s brother Manish, and built a device that did one thing with absolute focus. The first one was released in 1993 and the style is very Macintosh Platform language. I would not recommend that unit as it is very limited. The AlphaSmart 3000 arrived in 2000 wearing that iconic translucent bondi-blue shell that matched the first iMac and little too much like the eMate. It was never meant to be a computer. It was meant to be a writing instrument. Thousands were sold to schools around the world. Twenty-six years later it still is an amazing writing device. I bought a stack of six on eBay for about sixteen dollars each. They live in my bag the way a good pen used to. When I sit under the trees or at a picnic table with no signal, I open a file and the words come cleaner. I write some of my X Aricles and some of the ReadMultiplex articles on these machines. Later I connect the AlphaSmart to my computer or iPhone, dump the text, and the modern world takes over only when I am ready for it. The process is so freaking simple. Plug it in and it looks like a keyboard to the computer, than press (SEND) and it literally types out the text like you are typing it. It will likely be compatible decades into the future. I also use the AlphaSmart Neo2 as it is more powerful, ha, you can get more fonts on the screen. It is the rare piece of technology that feels more honest with every year that passes. No planned obsolescence. No forced updates. Just three batteries, a keyboard, and the freedom to think without interruption. Absolutely worth it. If you find one on eBay let me know. The best part is to see some kid’s homework still stuck in the machine from 2000.

Brian Roemmele

33,382 次观看 • 1 个月前

MIT DEDICATED A FULL LECTURE TO GIT'S INTERNALS -- BECAUSE THEY FOUND MOST DEVS MEMORIZE THE COMMANDS AND HAVE NO IDEA WHAT THE TOOL ACTUALLY DOES A whole 85 minutes MIT session that refuses to teach git as a list of commands to copy, and instead shows you the data model underneath -- the thing that makes every command finally make sense. -> The moment it clicks, git stops being scary magic. You stop memorizing "The incantation that fixed it last time" and start actually knowing what's happening. Most people learn just enough git to not get fired. Four commands, blind faith, and a prayer before every merge. In 2026 that's not enough anymore -> git is the literacy test for being in the room, and "I'll just reclone it" is the fastest way to look junior. An AI agent will branch, commit and rebase faster than you can read. When it tangles the history, untangling it runs on understanding the model MIT teaches in this one hour. Anyone can run git push. The person who understands the graph underneath is the one who saves the repo when it breaks. Bookmark & Watch it ↓
1:24:59

MIT DEDICATED A FULL LECTURE TO GIT'S INTERNALS -- BECAUSE THEY FOUND MOST DEVS MEMORIZE THE COMMANDS AND HAVE NO IDEA WHAT THE TOOL ACTUALLY DOES A whole 85 minutes MIT session that refuses to teach git as a list of commands to copy, and instead shows you the data model underneath -- the thing that makes every command finally make sense. -> The moment it clicks, git stops being scary magic. You stop memorizing "The incantation that fixed it last time" and start actually knowing what's happening. Most people learn just enough git to not get fired. Four commands, blind faith, and a prayer before every merge. In 2026 that's not enough anymore -> git is the literacy test for being in the room, and "I'll just reclone it" is the fastest way to look junior. An AI agent will branch, commit and rebase faster than you can read. When it tangles the history, untangling it runs on understanding the model MIT teaches in this one hour. Anyone can run git push. The person who understands the graph underneath is the one who saves the repo when it breaks. Bookmark & Watch it ↓

slash1s

459,113 次观看 • 2 个月前