'Keep AI on the leash' because it's far from perfect, says OpenAI's cofounder Andrej Karpathy
Andrej Karpathy thinks we're getting way too excited about AI, especially when it comes to deploying agents that act without supervision.
In a keynote at an event hosted by Y Combinator earlier this week, the computer scientist said people need to "keep AI on the leash." The OpenAI cofounder said current large language models still make mistakes no human ever would.
Karpathy likened LLMs to "people spirits" — uncanny simulations of human intelligence that hallucinate facts, lack self-knowledge, and suffer from "amnesia."
"They will insist that 9.11 is greater than 9.9 or that there are two R's in 'strawberry,'" Karpathy said in a talk published on Y Combinator's YouTube channel on Thursday. "They're going to be superhuman in some problem-solving domains and then they're going to make mistakes that basically no human will make."
Even though LLMs can churn out 10,000 lines of code in seconds, he said, that doesn't mean developers should sit back and let them run wild. "I'm still the bottleneck," he said. "I have to make sure this thing isn't introducing bugs."
"It gets way too overreactive," he added.
Karparthy urged developers to slow down and write more concrete prompts.
"I always go in small incremental chunks. I want to make sure that everything is good," he said.
"It makes a lot more sense to spend a bit more time to be more concrete in your prompts, which increases the probability of successful verification, and you can move forward," he added.
Karparthy did not respond to a request for comment from Business Insider.
The OpenAI cofounder coined the term "vibe coding" in February to describe the process of prompting AI to write code. The idea, he said, is that developers can "fully give in to the vibes" and "forget the code even exists."
AI still needs supervision
Karpathy isn't the only one urging caution.
Bob McGrew, OpenAI's former head of research, said on an episode of Sequoia Capital's "Training Data" podcast earlier this week that human engineers are still essential — not just to guide AI, but to step in when things get messy.
When something goes wrong or if a project "becomes too complicated for AI to understand," a human engineer can help break the problem down into parts for an AI to solve.
AI agents are like "genies," said Kent Beck, one of the authors of the seminal "Agile Manifesto" — they'll often grant your wish, but not always in the way you'd like them to.
"They will not do what you mean. They have their own agenda," Beck said on a recent episode of " The Pragmatic Engineer" podcast. "And the best analogy I could find is a genie. It grants you wishes, and then you wish for something, and then you get it, but it's not what you actually wanted."
Beck also said results are so inconsistent that using AI to code can sometimes feel like gambling.
Despite the nascent tech's limitations, even the biggest tech companies are betting on AI for the future of coding. AI writes more than 30% of Alphabet's new code, up from 25% last year, said CEO Sundar Pichai on the company's most recent earnings call.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
CNBC
15 minutes ago
- CNBC
Palantir co-founder Joe Lonsdale on Israel-Iran conflict, AI tech wars and future of DOGE
Joe Lonsdale, Palantir co-founder and 8VC founding partner, joins 'Squawk Box' to discuss the latest in the Israel-Iran conflict, what the role of America should be, impact of the 'bunker buster' bomb, AI tech wars, state of OpenAI-Microsoft relationship, future of DOGE, and more.
Tom's Guide
44 minutes ago
- Tom's Guide
ChatGPT Record quietly rolled out for Pro users — here's why I think free accounts could get voice messages soon
OpenAI has quietly rolled out a new 'Record' mode for ChatGPT — but for now, it's limited to Pro, Enterprise, and Edu users on the macOS desktop app. The new feature lets you tap a microphone icon and record a short voice message instead of typing. It's a small change, but one that enables ChatGPT to support users as a true voice assistant, making it faster and more conversational than ever. With Record mode, users simply speak their question, and ChatGPT will generate a response based on their audio input. A quick transcription of what you said also appears on screen, keeping the interaction clear and easy to follow. This is a briefer experience, not a full voice conversation like ChatGPT Voice, but it is designed for quick queries on the desktop. Currently limited to Mac users on the ChatGPT app — and only those on a Pro paid plan. Get instant access to breaking news, the hottest reviews, great deals and helpful tips. But I wouldn't be surprised to see it expand to mobile apps and free accounts in the near future. Why? It fits perfectly into OpenAI's larger strategy of making ChatGPT more multimodal; now it further combines voice, vision and text in one experience to support a seamless AI assistance. And let's not forget the competition: Google Gemini Live already supports real-time voice interaction across Android and iOS. If OpenAI wants ChatGPT to match that level of usability — especially on mobile — bringing Record mode to more users makes a lot of sense. Since OpenAI launched ChatGPT Voice for mobile, which I have tested and found incredibly useful, the next logical step is Record mode. However, ChatGPT Record is different, similar to AI transcription apps. It's fast, lightweight, and well-suited for quick interactions when you don't want a full back-and-forth conversation. With multimodal capabilities now a key battleground in the AI assistant space, I'd expect OpenAI to continue expanding features like Record. Giving free-tier users a taste of these tools helps build loyalty and could drive upgrades to paid plans. For now, if you're using ChatGPT Pro, Enterprise, or Edu on a Mac, look for the new mic icon next to the chat box to try Record mode. If you're not on a paid plan, keep an eye out; this is one feature that could be making its way to the broader ChatGPT user base sooner than we may think.
Yahoo
an hour ago
- Yahoo
SoftBank Plots $1 Trillion AI Hub With TSMC, Trump
SoftBank (SFTBY) is plotting a $1 trillion AI hub in Arizona alongside TSMC (NYSE:TSM) and the Trump administration to onshore advanced manufacturing, Bloomberg reported. The proposaldubbed Project Crystal Landwould mirror Shenzhen's industrial park model by hosting AI-powered robot assembly lines and high-tech chip fabrication facilities. Warning! GuruFocus has detected 7 Warning Signs with COIN. Son has held talks with Commerce Secretary Gina Raimondo on tax incentives for companies that set up shop there, and has quietly approached Samsung and Vision Fund portfolio firms about establishing factories within the complex. SoftBank's own Vision Fund led a $40 billion funding round in OpenAI earlier this year, signaling its appetite for deepening U.S. AI ties, and SFTBY shares are up 3.4% year-to-date, reflecting investor enthusiasm for the vision. Despite the fanfare, TSMC's exact role remains unconfirmed, and much hinges on federal sign-off under President Trump's industrial revival agenda and support from Arizona's state government. Son's pitch aligns with Washington's push to reduce reliance on overseas supply chains, but critics caution that building a trillion-dollar ecosystem from scratch could face hurdles from local regulators, labor shortages and capital intensity. Still, if SoftBank secures carve-outs on corporate taxes and streamlined permitting, Project Crystal Land could set a new benchmark for tech reshoring. Why it matters: Investors should watch for government backing and corporate commitments that will drive SoftBank's capital allocation and influence SFTBY's growth trajectory in the U.S. AI market. Closing: Market participants will be looking for formal announcements on tax-break packages and partnering agreements as Son's plan moves from pitch to planning. This article first appeared on GuruFocus.
