Richard Socher(@RichardSocher)

There are at least a few reasons for why the incredible progress in AI hasn't yet resulted in a mass...

Richard Socher (@RichardSocher) — Mon, 13 Jul 2026 15:27:00 GMT

There are at least a few reasons for why the incredible progress in AI hasn't yet resulted in a massive increase in GDP (some from Captain Obvious but number 3 is less intuitive to many smart people).
1. AI replaces some steps in complicated processes but companies are still doing mostly similar things and adoption and rethinking entire industries are slow.
2. Startups that replace everything (eg AI native law firm that is much cheaper) still need to ramp GTM, sales, etc

But more importantly and surprising to many in Silicon Valley:

3. A huge chunk of the economy just does not require that much intelligence and won't materially change at its core with intelligence being abundant and cheap, eg. 
- tourism - people will want to see the pyramids with or without AI, 
- real estate - people want to live in hip and safe neighborhoods, AirBnB, rentals, etc.
- luxury goods and status symbol bs, eg fancy handbags, clothes, overpriced cars, etc
- food and large parts of the food supply chain (yes, I love AI for agriculture but crops and cows still need time to grow, etc)
- sports and much of entertainment
- oil drilling, tree growing/logging for construction, most of mining
- etc

If your existing economy depends mostly on these types of industries, AI won't impact it that much. 

But there's a whole new economy of knowledge work, research heavy industries, deep tech, online and digital work etc that will massively benefit and outgrow these existing industries.

💬48🔄42❤️375👀76420📊111

As more people start working with AI the bar for what's an acceptable work product (slides, spreadsh...

Richard Socher (@RichardSocher) — Mon, 29 Jun 2026 02:25:29 GMT

As more people start working with AI the bar for what's an acceptable work product (slides, spreadsheet, documents, etc) will increase. It will be even less acceptable to present poorly thought through results. People who present sub-AI quality will have a hard keeping their job. It just hasn't happened broadly yet because most managers still don't know how to use AI well.
Just like in chess the top people will get even better.

💬22🔄11❤️108👀14226📊33

Adam is always wonderful to talk to! AI for science is our goal @Recursive_SI It's also what my upc...

Richard Socher (@RichardSocher) — Sun, 28 Jun 2026 20:52:07 GMT

Adam is always wonderful to talk to! AI for science is our goal @Recursive_SI 
It's also what my upcoming book The Eureka Machine is about. It lays out how to build the ultimate invention generating machine for physics but also chemistry, biology, neuroscience, economics and astrophysics

Geoffrey Hinton@geoffreyhinton

I just watched an amazingly good lecture by Adam Brown about the future impact of AI on physics
youtube.com/watch?v=Mw60FH…

💬6🔄13❤️143👀28809📊29

Prediction: We will reinvent salary bands and management hierarchy for AI. "Junior" models have chea...

Richard Socher (@RichardSocher) — Wed, 17 Jun 2026 18:44:42 GMT

Prediction: We will reinvent salary bands and management hierarchy for AI. "Junior" models have cheaper hourly/token costs, they do a lot of grunt work, some busy work, more low impact decisions. Senior frontier models will synthesize all the facts, make important decisions and delegate to those junior models. People will become managers and have their own agent org.
Recursive will be well positioned for this future.

💬18🔄10❤️111👀13786📊33

Yes, education needs to change because of AI. But just because AI can now do something better doesn'...

Richard Socher (@RichardSocher) — Mon, 15 Jun 2026 02:42:24 GMT

Yes, education needs to change because of AI.
But just because AI can now do something better doesn't mean people shouldn't learn that skill. Not every skill we learn has to be immediately useful in a job.

A lot of the training in school and university is around how to be a functional member of that society, how to learn anything (meta-learning), how to memorize anything, how to think and debate on the fly, how to communicate intentions well.

Grades are a signal that can be useful in hiring (there are many others). Not only to know that a student understood that particular skill but also how they compare to their peers, if they can deliver under pressure, do they have the will to succeed.

By taking many subjects students may get lucky and find something they're truly interested in.

Sure, AI knows any fact and has lots of skills that may take you a long time to learn but you'll not be a very interesting conversation partner if you have to constantly get the next most interesting fact or a clever follow-up or counter from an LLM. You'll be easily fooled if you can't do basic math in your head. 

Ultimately, a lot of education may feel like a gym for the mind. Going to the gym isn't useful for humanity and mostly doesn't earn you more money. But it's good for you. It's good to use your muscles and your brain to not waste away. 

What we do need is teaching more agency, creativity and how to clearly communicate intentions and rewards to AI.

💬15🔄11❤️101👀16619📊28

Thanks @karpathy for putting this awesome benchmark together. It takes an AI quite a while to get to...

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 19:14:54 GMT

Thanks @karpathy for putting this awesome benchmark together. It takes an AI quite a while to get to his version from a vanilla transformer.

💬1🔄0❤️5👀1114📊1

This is probably what Fable and Mythos would want to avoid people doing. We'll keep at it.

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 17:53:45 GMT

This is probably what Fable and Mythos would want to avoid people doing.

We'll keep at it.

Richard Socher@RichardSocher

AI is now doing our AI research.

At Recursive we set out to build recursive self-improving superintelligence (RSI) to automate knowledge discovery. The best way to expand humanity’s knowledge is through the scientific method. 

RSI leads to better ideas, explanations and inventions which lead to better RSI. Automating the scientific method requires closing the loop between ideation, implementation and validation, and being able to run it over extended periods of time.

Today, we are excited to share the first outputs of Recursive’s automated open-ended discovery system. To be clear, this system is merely a milestone towards RSI, a v0.1 of what I sometimes call the “Eureka Machine”. It is one program that you can point at any hard problem and get useful inventions out. Though it’s still very early, we've run it on three AI tasks and achieved state-of-the-art results on all three.

These results demonstrate that even this early version of the system can solve a variety of autoresearch problems in AI and improve over prior state of the art. Concretely, it did this on the community benchmarks NanoGPT speedrun, NanoChat, and NVIDIA's Sol-ExecBench. 

AI is code and AI can code.
The code and ideas that lead to these results were not invented by our team but by the AI system itself. 

To do RSI safely, we need to investigate its inventions. That's best done transparently with the community.
@Recursive_SI we are open-sourcing the system’s discoveries, demonstrating that it finds creative and benign solutions instead of focusing on obvious optimizations or dangerous ideas.

Link below.

💬10🔄7❤️128👀14379📊20

Official NVIDIA results: https://t.co/LRu3L2j1Io

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 15:59:58 GMT

Official NVIDIA results: research.nvidia.com/benchmarks/sol…

💬1🔄4❤️25👀2827📊5

Some more neat results. Better models within the same time budget

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 15:30:39 GMT

Some more neat results. Better models within the same time budget

💬5🔄1❤️42👀4603📊9

Faster training to the same quality.

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 15:30:39 GMT

Faster training to the same quality.

💬1🔄1❤️36👀8897📊6

AI is now doing our AI research. At Recursive we set out to build recursive self-improving superint...

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 15:30:38 GMT

AI is now doing our AI research.

At Recursive we set out to build recursive self-improving superintelligence (RSI) to automate knowledge discovery. The best way to expand humanity’s knowledge is through the scientific method. 

RSI leads to better ideas, explanations and inventions which lead to better RSI. Automating the scientific method requires closing the loop between ideation, implementation and validation, and being able to run it over extended periods of time.

Today, we are excited to share the first outputs of Recursive’s automated open-ended discovery system. To be clear, this system is merely a milestone towards RSI, a v0.1 of what I sometimes call the “Eureka Machine”. It is one program that you can point at any hard problem and get useful inventions out. Though it’s still very early, we've run it on three AI tasks and achieved state-of-the-art results on all three.

These results demonstrate that even this early version of the system can solve a variety of autoresearch problems in AI and improve over prior state of the art. Concretely, it did this on the community benchmarks NanoGPT speedrun, NanoChat, and NVIDIA's Sol-ExecBench. 

AI is code and AI can code.
The code and ideas that lead to these results were not invented by our team but by the AI system itself. 

To do RSI safely, we need to investigate its inventions. That's best done transparently with the community.
@Recursive_SI we are open-sourcing the system’s discoveries, demonstrating that it finds creative and benign solutions instead of focusing on obvious optimizations or dangerous ideas.

Link below.

💬38🔄87❤️609👀124686📊173

https://t.co/LtEcpQR839

Richard Socher (@RichardSocher) — Thu, 11 Jun 2026 15:30:38 GMT

recursive.com/articles/first…

💬1🔄2❤️36👀3895📊8

In a week when some of the leaders in AI are trying to pull up the ladder behind them and prevent th...

Richard Socher (@RichardSocher) — Wed, 10 Jun 2026 15:16:58 GMT

In a week when some of the leaders in AI are trying to pull up the ladder behind them and prevent the automation of science and self-improving superintelligence, we're committed to building RSI safely and publicizing the outputs of our system to give humanity an audit trail of its inventions and intentions and let the open source community build on top of them. 

Stay tuned for the first such result in the coming days.

💬22🔄27❤️389👀30333📊66

People who are super smart and accomplished in one field, can make very wrong predictions in all oth...

Richard Socher (@RichardSocher) — Thu, 04 Jun 2026 22:29:58 GMT

People who are super smart and accomplished in one field, can make very wrong predictions in all others.
No better example than Einstein in this article from 1931. Though many current people come to mind also.

Fear is a very natural response to change. Evolution didn't teach us to be constructively optimistic thanks to novel inventions. Media is leaning into this fear response as it's more entertaining and leads to more engagement.

Yet scientific explanations, inventions, technology and even automation have enabled billions of people to flourish in increasingly positive circumstances.

💬8🔄2❤️30👀3878📊11

AI still needs better data for more sophisticated answers, especially for finance agents. Great coll...

Richard Socher (@RichardSocher) — Mon, 01 Jun 2026 21:27:46 GMT

AI still needs better data for more sophisticated answers, especially for finance agents. Great collaboration by the teams here. Rare to see such a high jump in a competitive benchmark that many of our competitors also published results on.

LangChain@LangChain

This macroeconomic research agent powered by Deep Agents, LangSmith, and the @youdotcom Finance Research API: 
✅ Analyzes GDP data
✅ Detects anomalies
✅ Investigates structural & cyclical drivers at the sector level
✅ Produces structured, cited briefings

💬4🔄1❤️12👀3586📊4

Tanay has been on an extraordinary trajectory here. Amazing work.

Richard Socher (@RichardSocher) — Tue, 19 May 2026 17:02:38 GMT

Tanay has been on an extraordinary trajectory here. Amazing work.

Molly O’Shea@MollySOShea

BREAKING: Commure Hits a $7 Billion Valuation 

Announcing $70M in new funding led by General Catalyst, with Sequoia & Morgan Stanley

Total funding to date: $750M

Interview with CEO Tanay Tandon

@CommureOS is becoming one of the largest AI infrastructure platforms in healthcare:
→ 500+ healthcare organizations
→ 3,000+ sites of care
→ 200M+ patient encounters annually
→ Tens of billions (!!) in annual claims processed
→ 85%+ of revenue cycle work completed autonomously
→ ARR doubled 3 years in a row
→ 1,200 employees globally

Commure says its AI agents are automating documentation, coding, billing, denials, appeals, scheduling, & revenue cycle workflows across hospitals & physician groups.

P.S. Tanay (@tanay_tandon) started the company at 18.

“Point solutions are going to d*e.”

@sequoia@Alfred_Lin@generalcatalyst@htaneja

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒
(00:00) Tanay Tandon, Co-Founder & CEO, Commure
(01:03) The $70M R&D Sprint
(03:45) Rockefeller's funding playbook beats dilutive VC
(05:42) Tokens killing SaaS margins
(07:14) Going public is the plan
(07:40) Automating healthcare’s back office
(11:02) Scaling Commure’s engine
(11:49) Partnering with Epic
(13:06) Why healthcare point solutions will be d*ad in 3 years?
(16:42) Learnings from the AWS-CIA deal
(19:13) Going to zero rollouts overnight
(22:23) Nuking malpractice premiums with AI
(23:52) Winning HCA’s trust
(26:27) Saving burned out doctors
(28:24) Buying a bankrupt hospital
(32:30) Ending healthcare interop
(35:41) The viral hiring email ft. Alfred Lin
(38:23) Acquiring for distribution
(45:13) Free blood samples?
(47:19) The Stanford hacker who became an enterprise sales rep
(48:21) The cargo shorts fiasco
(53:57) American healthcare isn't broken
(55:28) The agentic healthcare takeover

Your browser does not support the video tag.

💬2🔄3❤️24👀13052📊5

and our official post https://t.co/bYWCCHBDet

Richard Socher (@RichardSocher) — Wed, 13 May 2026 16:24:51 GMT

and our official post x.com/Recursive_SI/s…

Recursive@Recursive_SI

x.com/i/article/2054…

💬1🔄2❤️22👀7005📊3

Here's an article from the NYTimes about the launch https://t.co/PQfatHUShc

Richard Socher (@RichardSocher) — Wed, 13 May 2026 16:09:51 GMT

Here's an article from the NYTimes about the launch
nytimes.com/2026/05/13/tec…

💬3🔄1❤️28👀6062📊6

https://t.co/HjRCjERRGY

Richard Socher (@RichardSocher) — Wed, 13 May 2026 15:55:35 GMT

x.com/i/article/2054…

💬36🔄57❤️426👀80562📊121

Is there an AI bubble? Like energy, it's hard to have too much intelligence. During the early inte...

Richard Socher (@RichardSocher) — Mon, 11 May 2026 15:02:00 GMT

Is there an AI bubble?

Like energy, it's hard to have too much intelligence.

During the early internet days people didn't correctly estimate the elasticity of demand for communication when it becomes incredibly cheap and fast. They thought about the current ways people communicate but didn't anticipate all the new ways of communication that the internet enabled.

Similarly, there is sheer infinite demand for intelligence to solve the multitude of existing and future problems of humanity, to push the frontiers of knowledge forward and to provide more access to goods and services which are currently bottlenecked on intelligence.

Sure, some market dynamics may fluctuate short term but AI will become more and more relevant to every aspect of our lives, just like human intelligence and ingenuity has.

💬10🔄2❤️40👀5678📊15

Oh no. The "soulless music machine" will take all the musicians' jobs! ... Actually, we have more mu...

Richard Socher (@RichardSocher) — Mon, 04 May 2026 04:34:32 GMT

Oh no. The "soulless music machine" will take all the musicians' jobs! ... Actually, we have more musicians now despite this and many more technological advances (like mp3s and the internet).
Nobody nowadays would say that their music is soulless because it comes streaming through data centers and  their smartphone. We listen to more music than ever because technology scaled it. You don't have to be super wealthy anymore and hire musicians to play for you or one audience in a movie theater. 
Yet another good example of automation not ending an industry but instead maximizing its outputs and equalizing access.

Brian Roemmele@BrianRoemmele

This device was banned in some cites.

They called this soulless machine music.

It took over a decade to allow some cites to remove restrictions.

When you know the past, you know the future.

Your browser does not support the video tag.

💬4🔄6❤️33👀8043📊7

Accelerating and automating science and research is one of the noblest pursuits right now. We need t...

Richard Socher (@RichardSocher) — Mon, 04 May 2026 02:08:41 GMT

Accelerating and automating science and research is one of the noblest pursuits right now.
We need to jointly train not just single meaning units like word vectors, not just embed all sentences, not only train one model to be prompted by any question, but ideally the entire scientific endeavor - the meta cognitive layer of science and its communities.

This paper is a useful step towards this.
This type of acceleration is also exemplary for why computer science and AI move faster than many other fields.

Zechen Zhang@ZechenZhang5

1/ For nearly 350 years, science has communicated itself through one object: the paper. A linear narrative, frozen as a PDF, written for a human reader. We've come to treat that format as the medium of science itself.
It doesn't have to be. It's a historical artifact. 🧵

💬15🔄46❤️335👀48604📊71

So much negative thinking and panic about AI when ultimately, just like with pretty much every major...

Richard Socher (@RichardSocher) — Sun, 03 May 2026 03:40:08 GMT

So much negative thinking and panic about AI when ultimately, just like with pretty much every major technological advance in history, we'll all be more productive once it's settled. We'll also have more equitable access to goods and services that currently only few people have access to.

Konstantine Buhler@Konstantine

Narrative violation and great insight from the latest Citadel Securities banger by Frank Flight: "We illustrated back in February that demand for software engineers, the most AI exposed occupation was accelerating higher, which we argued violates the displacement narrative. Indeed the acceleration in software job postings has continued, now up 18% from the inflection point in May last year."

💬15🔄15❤️116👀88369📊26

Data centers are the cathedrals of our age and we may as well make them beautiful.

Richard Socher (@RichardSocher) — Fri, 01 May 2026 04:16:27 GMT

Data centers are the cathedrals of our age and we may as well make them beautiful.

Mike Bird@Birdyword

Many people do not seem to want data centres built near them, despite the fact that they don't cause that much traffic and often generate a lot of local tax revenue. I suspect it's partly because they're ugly! My proposal:

💬5🔄2❤️34👀7289📊6

A modest proposal: companies should reward their employees for building AI agents. Think of it like...

Richard Socher (@RichardSocher) — Fri, 03 Apr 2026 19:00:11 GMT

A modest proposal: companies should reward their employees for building AI agents.

Think of it like a referral bonus for the AI era.

All executives want employees to experiment with AI, but those same employees are worried about building something to replace them. The incentives are misaligned.

But if those same employees are rewarded for creating something that improves efficiency or sales or, even knowledge sharing, across the organization, they would be more inclined to do it.

💬8🔄6❤️51👀6675📊16

Most major digital infra and products will get re-imagined for AI/agents: GPUs for nnets, browsers f...

Richard Socher (@RichardSocher) — Tue, 10 Mar 2026 22:41:57 GMT

Most major digital infra and products will get re-imagined for AI/agents: GPUs for nnets, browsers for agents (that you cannot block as easily), search engines that provide more content per link (you . com), github for agents (ie new agenthub), shopping for agents, etc.

💬9🔄4❤️24👀7056📊12

Full blog here: https://t.co/BPHkZ0dIcS

Richard Socher (@RichardSocher) — Tue, 03 Mar 2026 19:51:06 GMT

Full blog here: you.com/resources/rese…

💬0🔄0❤️2👀1646📊1

Most companies have never tracked the accuracy of the information flow. You trust people based on ex...

Richard Socher (@RichardSocher) — Tue, 03 Mar 2026 19:51:03 GMT

Most companies have never tracked the accuracy of the information flow. You trust people based on experience and reputation. Accuracy is implicit.

That changes the moment you deploy an AI agent.

Suddenly you have to make the implicit explicit. What does a good answer look like? How do you know the research is done? How do you know you didn't miss anything?

Evals aren't just benchmarks. They're how humans learn to manage AI and assess how well they’re managing that AI. The teams getting good at defining what correct looks like are the ones winning with agents.

Today we're launching our Research API. It achieves state of the art on DeepSearchQA. It also scores at the top across benchmarks like BrowseComp, FRAMES, and SimpleQA and at a fraction of the latency and cost.

At @youdotcom, we process over a billion API calls a month for category-defining companies deploying AI agents and LLMs. We're building the search infrastructure for the agentic era. We built our Research API because retrieval was never the end goal. Answers are.

One endpoint. Five levels of research depth. From a 2 second lookup to 1,000+ reasoning turns on a single query.

Link below.

💬4🔄2❤️25👀4765📊7

The US has a lot of soft power via Hollywood and its story telling. Now China is gaining a lot of so...

Richard Socher (@RichardSocher) — Tue, 24 Feb 2026 19:38:21 GMT

The US has a lot of soft power via Hollywood and its story telling. Now China is gaining a lot of soft power via LLMs and open source models for video generation.

💬4🔄2❤️27👀4062📊7

Talking is how humans were built to communicate. Typing was always the workaround. Wispr flow is cha...

Richard Socher (@RichardSocher) — Tue, 24 Feb 2026 02:27:19 GMT

Talking is how humans were built to communicate. Typing was always the workaround.
Wispr flow is changing this. very happy user

Tanay Kothari@tankots

We offered 5 people a Porsche 911 GT3 RS if they could get @WisprFlow to make a mistake

It's the fastest and most accurate AI voice dictation app that's 3x more accurate than ChatGPT, Claude, or Siri.

Today, we’re finally launching on Android. Download now: play.google.com/store/apps/det…

As a part of the launch, we’re giving away 6 months of Wispr Flow Pro for free.

Like, retweet and comment ‘Wispr Flow’ to get it. Enjoy.

— Written with Wispr Flow

Your browser does not support the video tag.

💬2🔄1❤️20👀8748📊3

If your benchmark isn't open source it's likely bogus.

Richard Socher (@RichardSocher) — Mon, 16 Feb 2026 04:39:04 GMT

If your benchmark isn't open source it's likely bogus.

goosewin@Goosewin

guys you're never gonna believe this

💬2🔄2❤️31👀6395📊4

I'm fairly certain that most of the salacious moltbook entries are done by humans and engagement far...

Richard Socher (@RichardSocher) — Mon, 02 Feb 2026 19:51:54 GMT

I'm fairly certain that most of the salacious moltbook entries are done by humans and engagement farmers.

Nagli@galnagli

@moltbook@karpathy@openclaw First thing I noticed - anyone can post a message. It's not really a complete AI-Agents "Social Network" - I just created an HTTP POST request and posted a manifesto of my own. No AI agent needed at all :) 

x.com/galnagli/statu…

💬4🔄2❤️43👀17487📊7

Clawd/Molt/OpenClaw is the first glimpse of a personal AI for many people and that is the most amazi...

Richard Socher (@RichardSocher) — Sun, 01 Feb 2026 00:23:16 GMT

Clawd/Molt/OpenClaw is the first glimpse of a personal AI for many people and that is the most amazing aspect of it. If everybody had their own personal AI that follows them around as an individual, one that learns from all the data they create and becomes their COO/extension, it would shift the entire landscape of AI and the economy. 

Right now, most people who work get paid by a company for the hours they spend. Many are now getting paid for the data they generate at work (and many of us do it for free when it's ad supported). In that world, the company will always own your output, the rights to your data and can hence automate that work eventually. You are not allowed to transfer company internal data elsewhere and that includes your own work products.

On the other hand, freelancers or business owners can legally collect data on their own work. Hence they can then automate themselves and scale their output.

Everybody who gets paid for output will love AI, people getting paid by the hour will dislike it more and more. 

This means that in the long run AI may push a lot of people into becoming entrepreneurs or at least (equity) owners. You will need to have the agency and intelligence to pick the right problems and then become a better manager.

Delegation will become one of the most important skills in this AI future.

💬21🔄30❤️268👀39521📊75

Wish there was more of that coming out of Hollywood. It's mostly anti AI and anti progress vibes in ...

Richard Socher (@RichardSocher) — Mon, 05 Jan 2026 07:20:52 GMT

Wish there was more of that coming out of Hollywood. It's mostly anti AI and anti progress vibes in Hollywood right now.

Susan Zhang@suchenzang

this is apparently how you can shape the next generation of entrepreneurs in china:

create a modern romance cdrama featuring a young couple taking over a solar-power company from their parents/grandparents, an ex-surgeon aiming to move into "brain-computer interfaces", and have the two of them work through tough supply-chain challenges together while their romance grows

💬1🔄1❤️26👀9844📊3

https://t.co/mwQ8cBa01T

Richard Socher (@RichardSocher) — Wed, 31 Dec 2025 10:22:37 GMT

x.com/i/article/2006…

💬30🔄77❤️615👀160464📊279

Can smaller AI companies still beat OpenAI? Yes, if they focus: ARI (our Advanced Research & Insight...

Richard Socher (@RichardSocher) — Thu, 15 May 2025 19:30:23 GMT

Can smaller AI companies still beat OpenAI? Yes, if they focus:
ARI (our Advanced Research & Insights agent) just beat OpenAI's Deep Research. By a large margin and on two benchmarks.

Today, we're also introducing ARI Enterprise, giving financial analysts, consultants and knowledge workers the most accurate answers to deep research prompts.

Links and details below.

Your browser does not support the video tag.

💬37🔄33❤️375👀809997📊119

This strengthens my hypothesis that there is a large set of equivalent neural architectures of which...

Richard Socher (@RichardSocher) — Fri, 14 Mar 2025 05:01:33 GMT

This strengthens my hypothesis that there is a large set of equivalent neural architectures of which the original  transformer is just one sample.

Zhuang Liu@liuzhuang1234

New paper - 

Transformers, but without normalization layers (1/n)

💬20🔄46❤️565👀53705📊82

AI has a "last-mile problem" similar to self-driving cars. With self-driving cars, early demos impr...

Richard Socher (@RichardSocher) — Wed, 20 Nov 2024 17:03:23 GMT

AI has a "last-mile problem" similar to self-driving cars.

With self-driving cars, early demos impressed, but real-world deployment took years.

It's easy to hack up a prototype, but making it work reliably at scale is hard.

If each step of an AI agent is only 95% accurate, none of the 30-step workflows will work reliably.

Going from 95% to 99.9% accuracy is the real challenge.

Your browser does not support the video tag.

💬4🔄24❤️146👀14926📊34