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.

I've built a new AI model that is not allowed to generate text. Instead, it produces a verifiable program that answers your prompt using the original reference. XERJ Reader-1 is a neurosymbolic model for working with data. It points to the exact words in your document, writes a small... show more

Ivan Novikov

7,649 subscribers

1,711,480 Aufrufe • vor 3 Tagen •via X (Twitter)

Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

63 Kommentare

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Classic transformer LLMs read well and compute badly, and code does the opposite. So XERJ Reader-1 only decides where to look and which of 56 symbols to apply. Exact code computes the answer and returns it with the words it came from. When the document has no answer, the program says so.

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Comparing XERJ Reader-1 and LLMs side by side makes the difference clear: XERJ Reader: marks where the answer lives in your document → writes a small program from those words → runs it and returns the answer LLM: reads the whole document → writes the answer token by token → a human checks it That's the whole model. It costs $10 per BILLION input tokens, 10x less than Claude Haiku 5.5 and 400x less than Claude Opus 5.5.

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Made-up quotes and citations are already costing companies and lawyers money: - Deloitte refunded A$97,587 to the Australian government for a report that quoted a federal judge saying something she never wrote - A US federal court in Oregon ordered two lawyers to pay $110,204 after their briefs cited 15 cases that don't exist (Couvrette v. Wisnovsky) I gave XERJ Reader-1 the court order Deloitte misquoted and asked who must establish that a benefits debt is owed, then where the judge says the line Deloitte quoted. Reader found no answer to either question in the document. The two checks cost 0.007¢ together (10x less than Claude Haiku).

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

How it works under the hood: Join the waitlist for early access: Talk tech with us on Discord: We can't wait to see what you throw at it.

Profilbild von Paul Mit

Paul Mitvor 3 Tagen

wow cool sounds very very helpful for product teams and tech founders! good luck with the product!

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

thanks!

Profilbild von SCOTTY BEAM

SCOTTY BEAMvor 3 Tagen

Congrats Ivan! 🤝 Huge milestone for you and the team. Really excited to see where you take this. Wishing you all the best and plenty more wins ahead!

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks so much! Really appreciate the support

Profilbild von Toby

Tobyvor 3 Tagen

Writing a formula over the exact words instead of an answer is such a clean idea. Every result is checkable.

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

You get it 🤝

Profilbild von bobaaa

bobaaavor 3 Tagen

looks intresting

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks a lot!

Profilbild von Kinder

Kindervor 3 Tagen

congrats on shipping this, genuinely one of the most interesting models i’ve seen in a while. excited to watch it grow

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Appreciate it! Plenty more to come 🙌

Profilbild von UGC Ninja

UGC Ninjavor 3 Tagen

wow claude is cooked

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Things are getting interesting 👀

Profilbild von Bobyhamster

Bobyhamstervor 3 Tagen

Verifiability over generated answers

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

That's what we're betting on. Answers you can actually check

Profilbild von wast3

wast3vor 3 Tagen

the most appealing thing is the price for the same result

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

That's a big part of what makes it exciting

Profilbild von 𝗗Λ𝗩𝗘 𓅓

𝗗Λ𝗩𝗘 𓅓vor 3 Tagen

really cool product gery helpful as well

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks! Glad you think so!

Profilbild von Gresham

Greshamvor 3 Tagen

nice approach

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Appreciate it!

Profilbild von Monreal

Monrealvor 3 Tagen

a model that’s not allowed to make things up is the most useful thing i’ve seen this month. congrats

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

That's a huge compliment, thank you!

Profilbild von Fluixo

Fluixovor 3 Tagen

verifiable programs with exact source pointers is the real unlock

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Exactly. That's what matters most

Profilbild von boan

boanvor 3 Tagen

thats a clever approach

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks!

Profilbild von Semichenko

Semichenkovor 3 Tagen

"output tokens free" got me. congrats on shipping this

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Appreciate it!

Profilbild von morph

morphvor 3 Tagen

before this I thought that Jev might be the breakthrough in AI tech. I was so freaking wrong.

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Means a lot! Excited for what's ahead

Profilbild von Fokki

Fokkivor 3 Tagen

so many advantages to use XERJ instead of Opus

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks! Excited to see what people build with it

Profilbild von Noir

Noirvor 3 Tagen

$0.01 per million input tokens and free output. For a pipeline that reads thousands of documents a day, that changes the budget.

Profilbild von Hans

Hansvor 3 Tagen

so it cites its sources, cute, still no idea if the answer's right

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Fair point. That's why XERJ Reader gives you more than a citation. You can actually rerun the program and see how it got to that answer

Profilbild von Sakata

Sakatavor 3 Tagen

Claude is cooked XERJ is way cheaper and faster

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Pretty wild difference, right?

Profilbild von maestro

maestrovor 3 Tagen

definitely a gamechanger

Profilbild von Lummox

Lummoxvor 3 Tagen

I thought Opus 5.5 is the best model

Profilbild von Ed

Edvor 3 Tagen

Wow. Verifiable answers instead of generated text could make data automation much more trustworthy.

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Couldn't agree more

Profilbild von Ed

Edvor 3 Tagen

Thank you for all the work you've done

Profilbild von slash1s

slash1svor 3 Tagen

that will be fr useful for me

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Glad to hear!

Profilbild von Diam

Diamvor 3 Tagen

I like that the calculation comes with the answer. Much easier to check an unexpected result when you can replay how it was produced.

Profilbild von Gipp 🦅

Gipp 🦅vor 3 Tagen

teams buried in recurring document questions should pay attention free output tokens change the automation budget!

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Exactly! Makes a huge difference for high-volume workloads

Profilbild von marfin

marfinvor 3 Tagen

congratulations on the release! this model is really a new round of development of the entire industry, it has never been so cheap and at the same time of high quality

Profilbild von Asteri

Asterivor 3 Tagen

a model that shows its work as code is exactly what data teams have been waiting for

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

That's exactly what we're aiming for!

Profilbild von palehonk

palehonkvor 3 Tagen

really great product

Profilbild von leopardracer

leopardracervor 3 Tagen

getting the program behind every answer means i can check the numbers myself, congrats on the launch

Profilbild von Trackmind

Trackmindvor 3 Tagen

Need to test this

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Looking forward to hearing your thoughts!

Profilbild von Bober_smart

Bober_smartvor 3 Tagen

The best deal on the market

Profilbild von AI Quanting

AI Quantingvor 3 Tagen

Picking where to look and which of 56 symbols to apply is still a judgement call. A wrong pick now ships with code attached, so it reads as authoritative. How does accuracy compare to a plain model on the same documents?

Profilbild von EugBass

EugBassvor 3 Tagen

wild thread man

Profilbild von Ivan Novikov

Ivan Novikovvor 3 Tagen

Thanks!

Profilbild von shiqway92

shiqway92vor 3 Tagen

actually impressive. faster, cheaper, and you can verify every answer

Ähnliche Videos

IN 1986 MIT FILMED THE LECTURE WHERE CODE STOPPED BEING CODE AND BECAME DATA 43 minutes from Gerald Sussman, in the most legendary intro programming course ever recorded. -> The idea that lands: to a program, your code is just data it can read and rewrite. He builds a program that does real calculus -- not by crunching numbers, but by reading the math as a list and transforming it piece by piece. Then he shows the trick hiding under all of it: in this language, code and data are the same material. Forty years later that is exactly what an AI coding agent is -- a program that reads your code as data, rewrites it, and hands it back. Sussman drew the whole idea on a blackboard in 1986. Writing code was never the deepest skill -- understanding that code itself is data a program can manipulate is. This is where you learn it. Most people think AI writing code is brand new. The ones who watch this saw the blueprint 40 years ago. Bookmark & Watch it. This one's a legend ↓

slash1s

22,107 Aufrufe • vor 4 Monaten

Data Models - Better AI Code in 5 Minutes Your AI isn't consistent because you're missing one file. I had the same problem until I started using a data model file. In this video, I'm showing you the exact two prompts I use: one generates the data model, the second uses it to generate features without breaking my app. Testing this with AnimeLeak (my actual production app) adding an admin approval feature. Using Droid with Sonnet 4.5 in spec mode, but this works in Cursor, Claude, or whatever you're using. KEY TIMESTAMPS 00:00 - The One File Your AI is Missing 00:44 - Prompt 1: Generate a Data Model 01:12 - The Result: Data Model Documentation 02:12 - Prompt 2: Implement a New Feature 03:00 - Reviewing the AI-Generated Spec 04:17 - Why This Workflow is So Powerful 04:32 - Your 3-Step Action Plan 05:12 - Next Video: Live Code Generation

Data Models - Better AI Code in 5 Minutes Your AI isn't consistent because you're missing one file. I had the same problem until I started using a data model file. In this video, I'm showing you the exact two prompts I use: one generates the data model, the second uses it to generate features without breaking my app. Testing this with AnimeLeak (my actual production app) adding an admin approval feature. Using Droid with Sonnet 4.5 in spec mode, but this works in Cursor, Claude, or whatever you're using. KEY TIMESTAMPS 00:00 - The One File Your AI is Missing 00:44 - Prompt 1: Generate a Data Model 01:12 - The Result: Data Model Documentation 02:12 - Prompt 2: Implement a New Feature 03:00 - Reviewing the AI-Generated Spec 04:17 - Why This Workflow is So Powerful 04:32 - Your 3-Step Action Plan 05:12 - Next Video: Live Code Generation

Ray Fernando

15,479 Aufrufe • vor 1 Jahr

introducing XERJ Corpus Hub, reference memory for AI agents. give your agent a corpus for your task. watch it become superintelligent. > finds the answers in indexed expert knowledge > applies them to your code, research, and projects > works with any model. free and OSS forever/Apache 2 install in one prompt: >Install XERJ (docs: index this project’s sources, docs, and logs, and connect as reference memory hub

introducing XERJ Corpus Hub, reference memory for AI agents. give your agent a corpus for your task. watch it become superintelligent. > finds the answers in indexed expert knowledge > applies them to your code, research, and projects > works with any model. free and OSS forever/Apache 2 install in one prompt: >Install XERJ (docs: index this project’s sources, docs, and logs, and connect as reference memory hub

Ivan Novikov

84,990 Aufrufe • vor 1 Monat

How can you solve complex tasks using a Large Language Model? Here is a 2-minute introduction to everything you need to know to 10x the quality of your results. Let's talk about three techniques, in order of complexity, starting with the easiest one: • In-Context Learning • Indexing + In-Context Learning • Fine-tuning In-Context Learning The team that trained GPT-3 found something they couldn't explain: You can condition a model using examples of how you want it to behave. I included an example prompt in the attached video. You can "teach" the model how you want it to interpret questions, select the correct answers, and format the results by giving a few examples. You can also give specific knowledge to the model that will be helpful when formulating answers. We call this approach "grounding the model." There's another example in the video. Indexing + In-Context Learning Unfortunately, there is a limit to how much data you can include in a prompt. We call this the "context size." One version of GPT-4 supports a context of approximately 6,000 words, while the other supports 25,000 words. Although this sounds like a lot, many applications need more than that. Imagine you wrote a book and want to build an application to answer any questions about your story. What happens if your book is longer than the context? That's where Indexing comes in. Using a model, you can turn every book passage into an embedding. These are vectors, numbers that "encode" the passage's text. You can then store these embeddings in a particular database that supports fast retrieval of these vectors. You can then turn any question into an embedding and search the database for the list of passages that are similar to that query. Instead of using the entire book to ask the model, you can now use the relevant passages as in-context information, effectively working around the context size limitation. Fine-tuning Fine-tuning can give you an extra boost to get reliable outputs from your LLM. It is, however, the most complex approach on the list. There are different approaches to fine-tuning a model with your data. A popular technique is to process your data with your LLM and use the outputs to train a new classifier that solves your specific task. Notice that here you aren't modifying the LLM. Instead, you are chaining it with your trained classifier. Another approach is to modify the parameters of the LLM using your data. Think of this as "rewiring" the model in a way that solves your particular task. The results and costs will vary depending on how many layers you want to fine-tune from the original model. Many companies think that fine-tuning is the solution to their problems. In my experience, many will benefit from exploring the other two approaches. I love explaining Machine Learning and Artificial Intelligence ideas. If you enjoy in-depth content like this, follow me Santiago so you don't miss what comes next.

How can you solve complex tasks using a Large Language Model? Here is a 2-minute introduction to everything you need to know to 10x the quality of your results. Let's talk about three techniques, in order of complexity, starting with the easiest one: • In-Context Learning • Indexing + In-Context Learning • Fine-tuning In-Context Learning The team that trained GPT-3 found something they couldn't explain: You can condition a model using examples of how you want it to behave. I included an example prompt in the attached video. You can "teach" the model how you want it to interpret questions, select the correct answers, and format the results by giving a few examples. You can also give specific knowledge to the model that will be helpful when formulating answers. We call this approach "grounding the model." There's another example in the video. Indexing + In-Context Learning Unfortunately, there is a limit to how much data you can include in a prompt. We call this the "context size." One version of GPT-4 supports a context of approximately 6,000 words, while the other supports 25,000 words. Although this sounds like a lot, many applications need more than that. Imagine you wrote a book and want to build an application to answer any questions about your story. What happens if your book is longer than the context? That's where Indexing comes in. Using a model, you can turn every book passage into an embedding. These are vectors, numbers that "encode" the passage's text. You can then store these embeddings in a particular database that supports fast retrieval of these vectors. You can then turn any question into an embedding and search the database for the list of passages that are similar to that query. Instead of using the entire book to ask the model, you can now use the relevant passages as in-context information, effectively working around the context size limitation. Fine-tuning Fine-tuning can give you an extra boost to get reliable outputs from your LLM. It is, however, the most complex approach on the list. There are different approaches to fine-tuning a model with your data. A popular technique is to process your data with your LLM and use the outputs to train a new classifier that solves your specific task. Notice that here you aren't modifying the LLM. Instead, you are chaining it with your trained classifier. Another approach is to modify the parameters of the LLM using your data. Think of this as "rewiring" the model in a way that solves your particular task. The results and costs will vary depending on how many layers you want to fine-tune from the original model. Many companies think that fine-tuning is the solution to their problems. In my experience, many will benefit from exploring the other two approaches. I love explaining Machine Learning and Artificial Intelligence ideas. If you enjoy in-depth content like this, follow me Santiago so you don't miss what comes next.

Santiago

384,573 Aufrufe • vor 3 Jahren

can AI write engaging news that people can trust? introducing ✨Data2Story: a data journalist agent. give it raw data, it generate a verifiable, multimodal article. 🔍verifiable: every claim is evidence-grounded, traces back to data, code, or a cited source. 🔮multimodal: the article is a generative UI — images, videos, audio, interactive charts. not just readable, but trustworthy and playable. 🧵1/N

can AI write engaging news that people can trust? introducing ✨Data2Story: a data journalist agent. give it raw data, it generate a verifiable, multimodal article. 🔍verifiable: every claim is evidence-grounded, traces back to data, code, or a cited source. 🔮multimodal: the article is a generative UI — images, videos, audio, interactive charts. not just readable, but trustworthy and playable. 🧵1/N

Kevin Lin

33,676 Aufrufe • vor 5 Monaten

Try this Git developer challenge: a bug appeared somewhere in the last 500 commits, and you have a test that detects it. What’s the shortest path to find the exact commit where it was introduced? Share your answer in the comments.

Try this Git developer challenge: a bug appeared somewhere in the last 500 commits, and you have a test that detects it. What’s the shortest path to find the exact commit where it was introduced? Share your answer in the comments.

Google for Developers

16,480 Aufrufe • vor 6 Monaten

Jev (join the waitlist on the homepage): Half your AI bill is one-word answers. Yes or no. Spam or not. Which bucket. You're paying a frontier model for that. Jev is built for exactly those jobs. You give it the options. It picks one in under half a second. It never writes a sentence. So it can't make stuff up. The answer has to be on your list. 444 times cheaper. 193 times faster. Like a light switch instead of a dimmer. It only has to do one thing. Hand it the boring jobs. Save the big models for real thinking. #jev #aimodel #llm #artificialintelligence #aitools

Jev (join the waitlist on the homepage): Half your AI bill is one-word answers. Yes or no. Spam or not. Which bucket. You're paying a frontier model for that. Jev is built for exactly those jobs. You give it the options. It picks one in under half a second. It never writes a sentence. So it can't make stuff up. The answer has to be on your list. 444 times cheaper. 193 times faster. Like a light switch instead of a dimmer. It only has to do one thing. Hand it the boring jobs. Save the big models for real thinking. #jev #aimodel #llm #artificialintelligence #aitools

jackroberts____

20,000 Aufrufe • vor 1 Monat

This is absolutely f*cking insane. A Stanford AI research group combines JEV with Claude Code to sort 78+ billion data points every 12 minutes. so the trick is stupidly simple: Jev runs a cheap first pass on everything. Claude only gets the hard cases. instead of: all data -> Claude -> $$$ it's: all data -> Jev -> only the unsure ones -> Claude and here's the part people miss. every Jev answer comes with a probability. that's your filter: -> Jev is 90%+ sure? keep the answer, done -> below that? send it up to Claude you set the cutoff once. the boring stuff never touches the expensive model. where this pays off right away: -> tagging support tickets -> sorting your inbox -> filtering leads -> cleaning a messy spreadsheet -> moderating comments faster. cheaper. way less compute and tokens burned. Jev sorts the mail so the genius only opens what matters. LLMs think. Jev decides. code does the job.

This is absolutely f*cking insane. A Stanford AI research group combines JEV with Claude Code to sort 78+ billion data points every 12 minutes. so the trick is stupidly simple: Jev runs a cheap first pass on everything. Claude only gets the hard cases. instead of: all data -> Claude -> $$$ it's: all data -> Jev -> only the unsure ones -> Claude and here's the part people miss. every Jev answer comes with a probability. that's your filter: -> Jev is 90%+ sure? keep the answer, done -> below that? send it up to Claude you set the cutoff once. the boring stuff never touches the expensive model. where this pays off right away: -> tagging support tickets -> sorting your inbox -> filtering leads -> cleaning a messy spreadsheet -> moderating comments faster. cheaper. way less compute and tokens burned. Jev sorts the mail so the genius only opens what matters. LLMs think. Jev decides. code does the job.

kiosa

18,441 Aufrufe • vor 27 Tagen

GPT-4o is bad at processing PDF documents. Whoever tells you otherwise is not living in the real world. In 2024, people fill out forms using pen and paper. Try to answer questions from those forms using modern models, and you'll be disappointed. I recorded a video to show you how to fix this. The answer is simple: stop letting the model see your PDF document. Instead, preprocess it and stick to showing the model text. Watch the video. You'll go from "THIS IS CRAP" to "EXCELLENT" in no time. I'm using Unstract to turn the documents into text while keeping the original format (this is crucial!) You can use them to process up to 100 pages for free. They collaborated with me on this post. You can find the code I wrote right here:

GPT-4o is bad at processing PDF documents. Whoever tells you otherwise is not living in the real world. In 2024, people fill out forms using pen and paper. Try to answer questions from those forms using modern models, and you'll be disappointed. I recorded a video to show you how to fix this. The answer is simple: stop letting the model see your PDF document. Instead, preprocess it and stick to showing the model text. Watch the video. You'll go from "THIS IS CRAP" to "EXCELLENT" in no time. I'm using Unstract to turn the documents into text while keeping the original format (this is crucial!) You can use them to process up to 100 pages for free. They collaborated with me on this post. You can find the code I wrote right here:

Santiago

196,882 Aufrufe • vor 2 Jahren

Hgraph AI Data Agent ✨ Now Live! Instantly chat with @Hedera data in your browser for free using our new data agent. Meet the network explorer that explores for you. Send a prompt → Watch now fast it is...

Hgraph AI Data Agent ✨ Now Live! Instantly chat with @Hedera data in your browser for free using our new data agent. Meet the network explorer that explores for you. Send a prompt → Watch now fast it is...

Hgraph

25,210 Aufrufe • vor 11 Monaten

AI models vs AI agents — the difference, in one picture People use "AI model" and "AI agent" like they mean the same thing. They're two completely different things — and once you see the line, the whole AI world snaps into focus. The MODEL is the brain. Under ChatGPT, Claude, and Gemini sits a model: it reads, writes, and reasons brilliantly — and it has one move. Words in, words out. It's a master chef with no kitchen: ask for dinner and you get a perfect recipe... but somebody still has to cook. The AGENT is that same chef with a kitchen built around him. Tools — real connections into your email, files, calendar — so the brain can actually do things. An order ticket instead of a question: a whole job. And the kitchen rhythm: cook, taste, adjust, until the dish is right. Ask a model about the emails you forgot to answer and it explains how to find them. Hand an agent that job and it opens the inbox, finds all seven, drafts every reply in your tone, and keeps going until each one is handled. A model hands you the recipe. An agent serves the meal. #aiagent #aiautomation #aiemployee

AI models vs AI agents — the difference, in one picture People use "AI model" and "AI agent" like they mean the same thing. They're two completely different things — and once you see the line, the whole AI world snaps into focus. The MODEL is the brain. Under ChatGPT, Claude, and Gemini sits a model: it reads, writes, and reasons brilliantly — and it has one move. Words in, words out. It's a master chef with no kitchen: ask for dinner and you get a perfect recipe... but somebody still has to cook. The AGENT is that same chef with a kitchen built around him. Tools — real connections into your email, files, calendar — so the brain can actually do things. An order ticket instead of a question: a whole job. And the kitchen rhythm: cook, taste, adjust, until the dish is right. Ask a model about the emails you forgot to answer and it explains how to find them. Hand an agent that job and it opens the inbox, finds all seven, drafts every reply in your tone, and keeps going until each one is handled. A model hands you the recipe. An agent serves the meal. #aiagent #aiautomation #aiemployee

Professor Glitch

47,600 Aufrufe • vor 2 Monaten

Larry Ellison just told every AI company on Earth they’re fighting the wrong war. The entire industry is racing to build the smartest model. More parameters. Better benchmarks. Faster inference. Ellison isn’t building a model. He’s controlling what every model needs to be useful. Every frontier AI trains on the same public internet. Same scraped pages. Same recycled text. When everyone has the same data, it’s not an advantage. It’s a floor. The only data that creates separation is private. Medical records. Financial models. Defense systems. Proprietary research locked behind firewalls for decades. That data already lives inside Oracle databases. Not Google’s. Not Microsoft’s. Not Amazon’s. Ellison didn’t enter the model war. He positioned himself above it. He rebuilt the database so AI can reason on private data without ever absorbing it. Training folds your data into the model permanently. Once it’s in, it never comes back out. Reasoning thinks with your data and hands back only the answer. The data never moves. One is surrender. The other is sovereignty. Ellison: “These are remarkable electronic brains.” He didn’t build the brain. He owns what the brain needs to think. Everyone is building the most powerful mind in human history. A mind is only as valuable as what it’s allowed to know. Own the knowledge and it doesn’t matter who builds the brain. That pattern has held through every era of human civilization. AI doesn’t break it. It proves it.

Larry Ellison just told every AI company on Earth they’re fighting the wrong war. The entire industry is racing to build the smartest model. More parameters. Better benchmarks. Faster inference. Ellison isn’t building a model. He’s controlling what every model needs to be useful. Every frontier AI trains on the same public internet. Same scraped pages. Same recycled text. When everyone has the same data, it’s not an advantage. It’s a floor. The only data that creates separation is private. Medical records. Financial models. Defense systems. Proprietary research locked behind firewalls for decades. That data already lives inside Oracle databases. Not Google’s. Not Microsoft’s. Not Amazon’s. Ellison didn’t enter the model war. He positioned himself above it. He rebuilt the database so AI can reason on private data without ever absorbing it. Training folds your data into the model permanently. Once it’s in, it never comes back out. Reasoning thinks with your data and hands back only the answer. The data never moves. One is surrender. The other is sovereignty. Ellison: “These are remarkable electronic brains.” He didn’t build the brain. He owns what the brain needs to think. Everyone is building the most powerful mind in human history. A mind is only as valuable as what it’s allowed to know. Own the knowledge and it doesn’t matter who builds the brain. That pattern has held through every era of human civilization. AI doesn’t break it. It proves it.

Dustin

98,131 Aufrufe • vor 4 Monaten

Fusion Models by robinhood:0xa92768863a55d8a0591709f7f5e594a249d36ea3, explained properly. Why a panel beats one model: Ask three models the same hard question and they disagree. The disagreement is the information, where they contradict each other is where one of them is wrong, and a judge can see it. OpenRouter shipped exactly this in June and their panels beat any single frontier model. Credit to them and their $10B sale to stripe. Why the same panel for everyone is the wrong shape: You don't have one best model. You have a best model per job. The one you trust for code is not the one you trust for writing, and neither is the one you'd pick for a five-line reply. A panel chosen per request, identical for every user, can't know that. A panel arranged around you can. The number that matters isn't price: It's cost per answer that works. Model A costs $4 per million tokens and gets your job right one time in four. Model B costs $10 and gets it right three times in four. Per answer you actually keep, A costs $16 and B costs $13. The cheap model was the expensive one. Everyone comparing price lists is measuring the wrong thing. You never rate anything: Stopping an answer halfway is a rating. Regenerating is a rating. Switching model and asking again is the loudest rating there is. Every one of those already writes a row we can learn from. The router learns which models earn your credits, per job, and tries a cheaper one now and then to check. Fusion only when it pays: Cost mode tries the cheap model first and escalates only if a judge says it failed. Performance mode fans out to your panel and merges. Most simple prompts never fuse. That's why it costs less, not more. Put it together and you have a personal model: Not trained. Assembled. Your profile from a two-minute questionnaire and everything you've imported from Claude or ChatGPT. Your memory, shared across every model so switching never starts from zero. Your skills, workflows you build once and run on any model. Your scorecard, which models earn your credits per job. One per person. Nobody else has yours. Every model is someone else's; the arrangement is you, and it leaves with you. People aren't loyal to models. They're loyal to results. A moat can't be a commodity, but it can be our own creation.

askr

13,924 Aufrufe • vor 1 Monat

Palantir CEO Alex Karp on why the model is not the moat. The application layer is: "To make them valuable in an enterprise, like a battlefield, regulated, or manufacturing context, you have to have what's called an application layer." "We have this thing called ontology that now everyone's copying. It takes a large language model and makes it safe and useful and precise." "Safe because it doesn't touch your underlying data." "Safe because it prevents the large language model from caching your data and replicating your business." The tell for any enterprise AI pitch: ask what sits between the model and your data. If the answer is nothing, you are not buying leverage. You are renting a fast way to hand your business to whoever owns the weights. Alex Karp, CEO of Palantir, on CNBC.

Palantir CEO Alex Karp on why the model is not the moat. The application layer is: "To make them valuable in an enterprise, like a battlefield, regulated, or manufacturing context, you have to have what's called an application layer." "We have this thing called ontology that now everyone's copying. It takes a large language model and makes it safe and useful and precise." "Safe because it doesn't touch your underlying data." "Safe because it prevents the large language model from caching your data and replicating your business." The tell for any enterprise AI pitch: ask what sits between the model and your data. If the answer is nothing, you are not buying leverage. You are renting a fast way to hand your business to whoever owns the weights. Alex Karp, CEO of Palantir, on CNBC.

Karl Mehta

31,891 Aufrufe • vor 4 Monaten

Anthropic's new model is extraordinary and it just revealed a problem that most enterprise AI buyers have not fully reckoned with yet (Save this), The model is genuinely impressive, and Chamath Palihapitiya assessment is that Anthropic continues to push the frontier harder than almost anyone. But that same update also showed their hand on something that changes the risk calculus for every business using Claude. Anthropic's new architecture stores every prompt you send for 30 days, no exceptions, not even for enterprise customers with zero-data retention agreements. The mechanism works like this, Anthropic now evaluates your prompt before generating output, deciding what it will and will not respond to, which means your query gets filtered before you even see a response. For individual users, that introduces a meaningful risk of censorship. For companies, Chamath says it is almost a non starter, and the reason is not just the data retention itself, it is the exposure that comes from operating at scale inside a large organization. A downstream scientist using the Claude APIs could accidentally trip a filter without knowing it, a business executive inside your company could trip it, and a molecular biology researcher could trip it and all of a sudden the company gets silently cut off from a tool it has embedded into critical workflows, with no warning and no recourse. Chamath gives Anthropic credit for being honest about how the system works, saying they tell the truth but notes that in this case the truth is not good. What this moment actually signals is a structural shift in how serious companies need to think about AI governance, because the question is no longer just which model performs best on benchmarks. It is who controls the model, who is learning from your data, and whether you are comfortable with a single point of failure sitting at the center of your competitive advantage. The answer for most enterprises will be broad model diversity, tighter governance frameworks and a serious reckoning with what it means to run mission-critical workflows through a third party that reserves the right to cut you off. Anthropic built a remarkable model and told the truth about how it works, the market's job now is to decide whether that transparency is enough to offset what the truth actually says.

Anthropic's new model is extraordinary and it just revealed a problem that most enterprise AI buyers have not fully reckoned with yet (Save this), The model is genuinely impressive, and Chamath Palihapitiya assessment is that Anthropic continues to push the frontier harder than almost anyone. But that same update also showed their hand on something that changes the risk calculus for every business using Claude. Anthropic's new architecture stores every prompt you send for 30 days, no exceptions, not even for enterprise customers with zero-data retention agreements. The mechanism works like this, Anthropic now evaluates your prompt before generating output, deciding what it will and will not respond to, which means your query gets filtered before you even see a response. For individual users, that introduces a meaningful risk of censorship. For companies, Chamath says it is almost a non starter, and the reason is not just the data retention itself, it is the exposure that comes from operating at scale inside a large organization. A downstream scientist using the Claude APIs could accidentally trip a filter without knowing it, a business executive inside your company could trip it, and a molecular biology researcher could trip it and all of a sudden the company gets silently cut off from a tool it has embedded into critical workflows, with no warning and no recourse. Chamath gives Anthropic credit for being honest about how the system works, saying they tell the truth but notes that in this case the truth is not good. What this moment actually signals is a structural shift in how serious companies need to think about AI governance, because the question is no longer just which model performs best on benchmarks. It is who controls the model, who is learning from your data, and whether you are comfortable with a single point of failure sitting at the center of your competitive advantage. The answer for most enterprises will be broad model diversity, tighter governance frameworks and a serious reckoning with what it means to run mission-critical workflows through a third party that reserves the right to cut you off. Anthropic built a remarkable model and told the truth about how it works, the market's job now is to decide whether that transparency is enough to offset what the truth actually says.

Milk Road AI

30,183 Aufrufe • vor 5 Monaten

Jev is cool. So is it's OSS companion, Laya. The Latest Cool Thing In AI™ tends to get a lot of hype, sometimes without everyone even understanding it. So... what is this thing? Jev is an AI model that consumes input and produces output VERY differently than chat, claude, grok. The input is two things: 1) Text state to assess. Email, html, code, whatever. 2) A set of questions which will be asked about the attached state. The canonical example from TypeSafe's docs is to identify the urgency of a support ticket. We pass the model the customer text + a single noul question "is this urgent?". Jev returns a full set of JSON. This JSON is not generated with token-by-token autoregression. Jev is not trained to produce sequences of text tokens, rather to answer questions, and guarantees well-formed responses. In the example below, we see it produces a 0.99 probability (on a 0-1.0 scale) that the answer is "yes." Jev supports exactly three types of questions (seconds example in video): a) Noul: 0–1 probability that the answer to a yes/no question is "yes." b) Choice: Ask question with pre-defined set of answers. Jev chooses the best and assigns probabilities to each. c) Score: Ask question with pre-defined scale of answers. Jev produces a position on the scale. Jev computes answers for all questions in parallel, making responses super fast even for many questions in a single request. This might seem like a narrow set of capabilities, but in the right contexts leads to incredible potential. It also makes for a useful API / primitive for programming, since the outputs are... *ahem*... type-safe and predictable in structure. Jev is not going to replace LLMs for writing your code, auto-generating your docs, or being at the core of an agent harness. But Jev IS incredibly cool, and will be used to build a lot of amazing tech. Hope this helps.

Jev is cool. So is it's OSS companion, Laya. The Latest Cool Thing In AI™ tends to get a lot of hype, sometimes without everyone even understanding it. So... what is this thing? Jev is an AI model that consumes input and produces output VERY differently than chat, claude, grok. The input is two things: 1) Text state to assess. Email, html, code, whatever. 2) A set of questions which will be asked about the attached state. The canonical example from TypeSafe's docs is to identify the urgency of a support ticket. We pass the model the customer text + a single noul question "is this urgent?". Jev returns a full set of JSON. This JSON is not generated with token-by-token autoregression. Jev is not trained to produce sequences of text tokens, rather to answer questions, and guarantees well-formed responses. In the example below, we see it produces a 0.99 probability (on a 0-1.0 scale) that the answer is "yes." Jev supports exactly three types of questions (seconds example in video): a) Noul: 0–1 probability that the answer to a yes/no question is "yes." b) Choice: Ask question with pre-defined set of answers. Jev chooses the best and assigns probabilities to each. c) Score: Ask question with pre-defined scale of answers. Jev produces a position on the scale. Jev computes answers for all questions in parallel, making responses super fast even for many questions in a single request. This might seem like a narrow set of capabilities, but in the right contexts leads to incredible potential. It also makes for a useful API / primitive for programming, since the outputs are... ahem... type-safe and predictable in structure. Jev is not going to replace LLMs for writing your code, auto-generating your docs, or being at the core of an agent harness. But Jev IS incredibly cool, and will be used to build a lot of amazing tech. Hope this helps.

Ben Dicken

42,183 Aufrufe • vor 1 Monat

I built a Reddit marketing agent that gets your business mentioned by ChatGPT, Gemini and Claude. Not a spam bot. Not the kind that gets banned by mods. An agent that drafts genuinely valuable answers grounded in your company's actual content and slips in a brand mention only when it actually helps. Google is paying Reddit $60M a year just for their data!!! If your business isn't showing up in that data, your competitors are, and ChatGPT is recommending them instead of you. Here's how the system works: → A Claude routine pulls the target subreddit's JSON endpoint every hour for new posts → A Phase 1 checklist filters every post — is it answerable from our wiki, factual, on-brand, mod-safe? → The answer engine queries an LLM Wiki built from our blog, guides, and help center (Firecrawl ingested every page into a raw → wiki → index structure) → Claude drafts a casual, value-first answer that grounds every claim in our content → Every draft lands in Slack for human-in-the-loop review before it posts — no fire-and-forget, no bans I built this in a week for CoinLedger, the crypto tax software I've been running for 8 years. 10% of our new signups already say they found us through ChatGPT. This system is how we push that number a lot higher. If you want the exact Claude skill I built: Like & Comment "AGENT" (must be following so I can dm you) I'll send it over for free.

David Roberts

36,737 Aufrufe • vor 6 Monaten

As an oncologist, I often see a patient right after their CT scan, and the scan hasn't been read yet. Reads can take a day or two. So I built a small tool to see what an open model can do. You ask a question about a chest CT in plain words, and it shows you the answer and the exact spot it looked at. I typed "find the nodules in the lungs" and it found and boxed the nodule. It can also compare two scans, which is what I care about most. As an oncologist I want to know: is it growing, shrinking, or stable? It's a research demo on a 4 billion parameter open model. Not a medical device. I haven't tested sensitivity or specificity here, and I still want a radiologist's eyes on every scan. But for a quick look at new nodules or progression in clinic, this is promising. (The video is trimmed: the model took about 10 seconds to answer.)

As an oncologist, I often see a patient right after their CT scan, and the scan hasn't been read yet. Reads can take a day or two. So I built a small tool to see what an open model can do. You ask a question about a chest CT in plain words, and it shows you the answer and the exact spot it looked at. I typed "find the nodules in the lungs" and it found and boxed the nodule. It can also compare two scans, which is what I care about most. As an oncologist I want to know: is it growing, shrinking, or stable? It's a research demo on a 4 billion parameter open model. Not a medical device. I haven't tested sensitivity or specificity here, and I still want a radiologist's eyes on every scan. But for a quick look at new nodules or progression in clinic, this is promising. (The video is trimmed: the model took about 10 seconds to answer.)

Roupen Odabashian

67,166 Aufrufe • vor 15 Tagen

$Big pharma just handed the AI industry one of the most important reality checks of 2026 (Save this). david friedberg revealed that Anthropic approached major life sciences companies with a pitch, share your proprietary data, sign an NDA and we will give you early access to a specialized life sciences model and nearly every company they spoke with said no. Here is what these pharma companies understood that many enterprises still have not. A large pharmaceutical company may have spent decades and tens of billions of dollars generating proprietary datasets, clinical trial results, genomic sequences, drug interaction data, compound libraries. That data is the business and the competitive moat that separates them from every other player in the industry lives in those datasets. Handing it to an AI lab in exchange for early access to a model is essentially handing your most valuable asset to a company whose entire business model depends on combining your data with everyone else's and then selling the output back to you and to your competitors. Palantir CEO Alex Karp made this exact point that enterprise leaders are paying for AI tokens that generate no tangible business value while simultaneously surrendering their most sensitive operational data to external providers. He called this transferring a company's alpha, the unique advantage that secures the business directly to a third-party lab. Microsoft CEO Satya Nadella echoed the same concern independently, warning that entire sectors might find their accumulated knowledge commoditized if they do not build their own data and model ownership layers. The structural problem is not unique to pharma but it applies to every enterprise sector. Every time an employee runs a query through a third-party frontier model, proprietary workflows, customer data, and strategic processes pass through infrastructure the enterprise does not control. The data already shows the market moving, Open-source captured 67% of all AI tokens processed in the first half of 2026, up from a fraction of that just twelve months earlier. The performance gap between proprietary frontier models and open-source alternatives has nearly closed, DeepSeek costs approximately 1/36th of GPT-5 for comparable workloads. What pharma figured out and what enterprises across every sector are starting to realize is that the model is not the moat but the data is. And once you hand your data to a model company, you have permanently surrendered the asset that took you decades and billions of dollars to build.$

Big pharma just handed the AI industry one of the most important reality checks of 2026 (Save this). david friedberg revealed that Anthropic approached major life sciences companies with a pitch, share your proprietary data, sign an NDA and we will give you early access to a specialized life sciences model and nearly every company they spoke with said no. Here is what these pharma companies understood that many enterprises still have not. A large pharmaceutical company may have spent decades and tens of billions of dollars generating proprietary datasets, clinical trial results, genomic sequences, drug interaction data, compound libraries. That data is the business and the competitive moat that separates them from every other player in the industry lives in those datasets. Handing it to an AI lab in exchange for early access to a model is essentially handing your most valuable asset to a company whose entire business model depends on combining your data with everyone else's and then selling the output back to you and to your competitors. Palantir CEO Alex Karp made this exact point that enterprise leaders are paying for AI tokens that generate no tangible business value while simultaneously surrendering their most sensitive operational data to external providers. He called this transferring a company's alpha, the unique advantage that secures the business directly to a third-party lab. Microsoft CEO Satya Nadella echoed the same concern independently, warning that entire sectors might find their accumulated knowledge commoditized if they do not build their own data and model ownership layers. The structural problem is not unique to pharma but it applies to every enterprise sector. Every time an employee runs a query through a third-party frontier model, proprietary workflows, customer data, and strategic processes pass through infrastructure the enterprise does not control. The data already shows the market moving, Open-source captured 67% of all AI tokens processed in the first half of 2026, up from a fraction of that just twelve months earlier. The performance gap between proprietary frontier models and open-source alternatives has nearly closed, DeepSeek costs approximately 1/36th of GPT-5 for comparable workloads. What pharma figured out and what enterprises across every sector are starting to realize is that the model is not the moat but the data is. And once you hand your data to a model company, you have permanently surrendered the asset that took you decades and billions of dollars to build.

Milk Road AI

16,317 Aufrufe • vor 4 Monaten