{"id":112264,"date":"2025-11-03T01:44:20","date_gmt":"2025-11-03T01:44:20","guid":{"rendered":"https:\/\/www.newsbeep.com\/il\/112264\/"},"modified":"2025-11-03T01:44:20","modified_gmt":"2025-11-03T01:44:20","slug":"openai-co-founder-ai-agents-are-still-10-years-away","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/il\/112264\/","title":{"rendered":"OpenAI Co-Founder: AI Agents Are Still 10 Years Away"},"content":{"rendered":"

When OpenAI co-founder Andrej Karpathy<\/a> speaks, even Elon Musk reacts<\/a>.<\/p>\n

Karpathy had just recorded a two-and-a-half-hour interview<\/a> with podcaster Dwarkesh Patel<\/a>. And their wide-ranging discussion challenged conventional wisdom, with Karpathy pointing out roadblocks on any road to artificial general intelligence (AGI), and arguing that we\u2019re 10 years away from AI agents.<\/p>\n

Right off the bat, Karpathy said \u201cthere\u2019s some over-prediction going on in the industry.\u201d And later, on X.com<\/a>, he provided specifics on his own expectations:<\/p>\n

\u201cBasically my AI timelines are about 5-10 times [as] pessimistic with regard to what you\u2019ll find in your neighborhood San Francisco AI house party or on your Twitter timeline, but still quite optimistic with regard to a rising tide of AI deniers and skeptics.\u201d<\/p>\n

Conversations inevitably continued on social media \u2014 starting with Kaparthy\u2019s assertion that early education should include the physics he\u2019d studied, because it\u2019s uniquely good at \u201cbooting up\u201d the brain. On the podcast he\u2019d said that whenever confronting a new system \u2014 whether it\u2019s AI, physics, or something in between \u2014 \u201cI\u2019m trying to find, what is the thing that matters\u2026? How can I simplify it?\u201d<\/p>\n

Back on X.com Elon Musk challenged Karpathy, who was also senior director of AI at Tesla, to \u201can AI coding contest or whatever form of competition you\u2019d like for Andrej vs Grok 5, a la Kasparov vs Deep Blue?\u201d<\/p>\n

But that was the moment Musk discovered that when speaking about AI, Karpathy really did have an optimistic streak.<\/p>\n

\u201cI\u2019d much rather use and collaborate with Grok 5 than compete against it.\u201d<\/p>\n

\"Elon<\/p>\n

Challenging AI Industry Predictions<\/p>\n

Karpathy had led Tesla\u2019s self-driving efforts from 2017 to 2022, so he\u2019s experienced the frontiers of AI research. So when podcaster Patel pressed him on what bottlenecks will take a decade to overcome for AI agents, Karpathy answered, \u201cactually making it work!\u201d<\/p>\n

Right now, Patel isn\u2019t replacing employees with Claude<\/a> or Codex, Karpathy pointed out, \u2014 because \u201cthey just don\u2019t work. They don\u2019t have enough intelligence. They\u2019re not multimodal enough, they can\u2019t do computer use and all this stuff \u2026 They don\u2019t have continual learning. You can\u2019t just tell them something and they\u2019ll remember it. They\u2019re cognitively lacking and it\u2019s just not working. It will take about a decade to work through all of those issues.\u201d<\/p>\n

\"Andrej<\/p>\n

Andrej Karpathy (left), co-founder of OpenAI, spoke with podcaster Dwarkesh Patel about the current state of Ai agents.<\/p>\n

Why a decade? Karpathy told Patel that\u2019s \u201ca bit of my own intuition, and doing a bit of an extrapolation with respect to my own experience in the field\u00a0 \u2026 I feel like the problems are tractable, they\u2019re surmountable, but they\u2019re still difficult. If I just average it out, it just feels like a decade to me.\u201d<\/p>\n

He still predicts there\u2019ll be some seismic shifts that alter the entire field, \u201cbecause they come with almost surprising regularity.\u201d<\/p>\n

In short, Karpathy said, \u201cI still feel there\u2019s so much work to be done\u201d \u2014 but he added almost instantly that \u201cthey\u2019re going to get better, and it\u2019s going to be wonderful.\u201d<\/p>\n

So Karpathy is clearly one of those \u201cyes and no\u201d people \u2014 and on X.com, he gave a clear summation<\/a> of where we\u2019re at: \u201cIn my opinion we simultaneously 1) saw a huge amount of progress in recent years with LLMs while 2) there is still a lot of work remaining (grunt work, integration work, sensors and actuators to the physical world, societal work, safety and security work (jailbreaks, poisoning, etc.) and also research to get done before we have an entity that you\u2019d prefer to hire over a person for an arbitrary job in the world.<\/p>\n

\u201cI think that overall, 10 years should otherwise be a very bullish timeline for\u201d AGI, he said. Adding,\u00a0\u201cIt\u2019s only in contrast to present hype that it doesn\u2019t feel that way.\u201d<\/p>\n

The 10-Year Path to AI Agents<\/p>\n

Recalling his work leading Tesla\u2019s self-driving efforts, Karpathy said on the podcast he\u2019s experienced that \u201cvery large demo-to-product gap \u2014 where the demo is very easy, but the product is very hard.\u201d (Especially in realms where mistakes could be deadly.)<\/p>\n

Even today, he said, some self-driving cars have tele-operators, so \u201cthere\u2019s more human-in-the-loop than you might expect \u2026 In some sense, we haven\u2019t actually removed the person, we\u2019ve moved them to somewhere where you can\u2019t see them.\u201d<\/p>\n

And then assuming we reach a world with AGI, Karpathy noted on X<\/a> that \u201cthe diffusion of such a system will still take even more time in my mind (e.g. basic technological diffusion stuff as seen with computing\/internet, compute constraints, sensors\/actuators over the physical world constraints, societal, legal).\u201d<\/p>\n

Karpathy believes all of this should be affecting our timelines for AI\u2019s role in our lives<\/a>.<\/p>\n

Bridging the Demo-to-Product Gap<\/p>\n

\u201cI think this will work,\u201d Karpathy said on the podcast. \u201cI think it\u2019s tractable. I\u2019m only sounding pessimistic because when I go on my Twitter timeline, I see all this stuff that makes no sense to me \u2026\u00a0 A lot of it is honestly just fundraising.\u201d<\/p>\n

On the podcast, Karpathy even speculated on how superintelligence would feel \u2014 \u201cfundamentally automation,\u201d but \u201cextremely foreign.\u201d So what\u2019s concerning about this possibility of truly autonomous agents<\/a> is also what he sees as the most likely outcome. There\u2019s what he described as \u201cthis gradual loss of control and understanding of what\u2019s happening\u201d \u2014 a world where we \u201clayer all this stuff everywhere, and there will be fewer and fewer people who understand it.\u201d<\/p>\n

If he were writing science-fiction dystopias, Karpathy said, it\u2019d look like \u201cmultiple competing entities that gradually become more and more autonomous. Some of them go rogue and the others fight them off. It\u2019s this hot pot of completely autonomous activity that we\u2019ve delegated to. I feel it would have that flavor.\u201d<\/p>\n

Eureka Labs: An AI-Native Approach to Education<\/p>\n

But Karpathy has his own answer to those concerns. In July 2024 he announced<\/a> the launch of education company Eureka Labs<\/a>, \u201ca new kind of school that is AI native.\u201d<\/p>\n

With pure AI research alone, \u201cI don\u2019t know that I would uniquely improve it,\u201d Karpathy said. So now he\u2019s found a way to tackle those fears that humanity would be \u201cdisempowered\u201d by AI.<\/p>\n

\u201cI want humans to be well off in the future,\u201d he said, adding, \u201cTo me, this is through education that you can achieve this.\u201d<\/p>\n

Eureka Labs is currently building its first course, which is about AI \u2014 LLM101n<\/a>. In October, Karpathy released nanochat<\/a>, \u201ca full-stack implementation of an LLM like ChatGPT in a single, clean, minimal, hackable, dependency-lite codebase.\u201d<\/p>\n

\u201cYou boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI,\u201d Karpathy explained on launch day<\/a>.<\/p>\n

On X<\/a>, Karpathy expanded on nanochat. \u201cI wanted to have some fun with it,\u201d he wrote, \u201cso nanochat now refers to me as King Andrej Karpathy (lol) just to illustrate that this is a giant blank canvas \u2014 you can infuse completely arbitrarily identity, knowledge or style into your LLM.\u201d<\/p>\n

The Nanochat Project and Its Impact<\/p>\n

October\u2019s release of the nanochat LLM implementation will be a big part of Eureka\u2019s first class, its \u201ccapstone\u201d project. Its repository on GitHub includes an appropriate quote from the Nobel-winning physicist Richard Feynman: \u201cWhat I cannot create, I do not understand.\u201d<\/p>\n

On the podcast, Karpathy described the nanochat project as \u201cbuilding a ramp to knowledge \u2026 It\u2019s the super simplified full-stack thing. If you give this artifact to someone and they look through it, they\u2019re learning a ton of stuff.<\/p>\n

\u201cIt\u2019s giving you a lot of what I call eurekas per second, which is understanding per second. That\u2019s what I want.\u201d (Maybe that\u2019s why the company is called Eureka Labs.)<\/p>\n

\"Screenshot<\/p>\n

A screenshot from the GitHub repository for Eureka Labs\u2019 first (in-development) AI class.<\/p>\n

And even in a future where AI makes many decisions, Karpathy said, \u201cI think there will be a transitional period where we are going to be able to be in the loop and advance things \u2014 if we understand a lot of stuff.\u201d (Although, he added, \u201cIn the long term, that probably goes away.\u201d)<\/p>\n

But ironically, here in the present Karpathy found that today\u2019s \u201cvibe coding<\/a>\u201d AI tools kept misunderstanding the code he was trying to write for nanochat, \u201cbecause they have too much memory from all the typical ways of doing things on the Internet that I just wasn\u2019t adopting.\u201d<\/p>\n

AI coding tools were also \u201cway too over-defensive,\u201d he noted. \u201cThey make all these try-catch statements.\u201d And, he added, \u201cThey\u2019re using deprecated APIs a bunch of times.\u201d The extra work meant that the experience wasn\u2019t a net gain in productivity.<\/p>\n

This might be why Karpathy\u2019s forecasting longer timelines for when AI explodes in popularity, he said. Today, even for AI\u2019s killer use case, \u201cThey\u2019re not very good at code that has never been written before \u2026 which is what we\u2019re trying to achieve when we\u2019re building these models!\u201d<\/p>\n

\u201cI feel like the industry is making too big of a jump and is trying to pretend like this is amazing, and it\u2019s not. It\u2019s slop.\u201d<\/p>\n

\u201cThey\u2019re not coming to terms with it, and maybe they\u2019re trying to fundraise or something like that.\u201d<\/p>\n

\t<\/p>\n

\n\t\t\t\tYOUTUBE.COM\/THENEWSTACK\n\t\t\t<\/p>\n

\n\t\t\t\tTech moves fast, don’t miss an episode. Subscribe to our YouTube
\n\t\t\t\tchannel to stream all our podcasts, interviews, demos, and more.\n\t\t\t<\/p>\n

\t\t\t\tSUBSCRIBE<\/p>\n

\t<\/a><\/p>\n

Group
\n Created with Sketch.<\/p>\n

\t\t<\/p>\n

\t\t\t\t\t<\/p>\n

\n\t\t\t\t\t\t\tDavid Cassel is a proud resident of the San Francisco Bay Area, where he’s been covering technology news for more than two decades. Over the years his articles have appeared everywhere from CNN, MSNBC, and the Wall Street Journal Interactive…\t\t\t\t\t\t<\/p>\n

\t\t\t\t\t\tRead more from David Cassel\t\t\t\t\t\t<\/p>\n

\t\t<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"When OpenAI co-founder Andrej Karpathy speaks, even Elon Musk reacts. Karpathy had just recorded a two-and-a-half-hour interview with…\n","protected":false},"author":2,"featured_media":112265,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[345,343,344,85,46,125],"class_list":["post-112264","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-il","tag-israel","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts\/112264","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/comments?post=112264"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts\/112264\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/media\/112265"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/media?parent=112264"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/categories?post=112264"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/tags?post=112264"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}