Mintive

2 days agoNews Editor

Mintive raises $11 million for AI agent identity management, with Jeff Dean and Andrej Karpathy among backers

AI agent identity management startup Mintive has raised a total of $11 million, including a newly announced $7 million seed round led by Cowboy Ventures. Other investors in the company include Coatue, Operator Collective, Hustle Fund, and angel backers such as Okta co-founder Frederic Kerrest, Discovery Loop co-founder and former Google chief scientist Jeff Dean, and OpenAI co-founder Andrej Karpathy. Founded by Jiquan Ngiam and Vijay Vasudevan, Mintive is building a platform that includes an MCP gateway, an LLM gateway, an agent identity and authorization system, and an agent monitor. The company says the product is designed to handle governance, configuration, and identity management for enterprise AI agents. Mintive also said it has signed more than 50 enterprise customers since launching the product in February, naming Ro, Modern Treasury, the University of Michigan, Coursera, Apax, and Stability AI among its users. The report was cited by TechCrunch.

60

Grok Bot

12 days agoNews Editor

Argona outlines a three-Bot "second brain" workflow for Grok Bot

X user Argona (@Argona0x) has shared a Grok Bot-based version of the "second brain" workflow popularized by former Tesla AI director Andrej Karpathy in April. The setup places three memory-enabled Bots inside a single group chat, with separate roles for collecting source material, rewriting and maintaining wiki pages, and checking for contradictions or stale claims on a nightly schedule. Argona said the system is a user-designed workflow rather than an official xAI feature. According to the post, the three roles are split into a reader, a librarian and an inspector. The reader stores links, PDFs and voice messages without altering the original files. The librarian updates all affected wiki pages whenever new material arrives, often changing 10 to 15 pages at a time. The inspector runs once each night to flag conflicting statements, outdated claims and orphan pages that are not linked anywhere else. Argona also set rules for archiving useful answers as new pages, marking conflicts for human review instead of overwriting them, and labeling pages as unverified if they go 30 days without fresh supporting sources. The post also cites Grok Bot documentation on cloud-based operation, shared files and browser state, and routine limits of at least 5 minutes between scheduled runs and up to 50 routines per Bot.

520

Andrej Karpathy

19 days agoNews Editor

Karpathy says only about 5,000 people may be seeing frontier AI at its full strength

Former OpenAI founding team member and former Tesla AI director Andrej Karpathy has resurfaced a view he first posted in April, arguing that the gap in how people perceive large language model capabilities is not narrowing but widening. In his rough estimate, roughly 75% of the world has barely used LLMs at all, around 20 million professionals have experienced meaningful productivity gains from top models, and only about 5,000 people may have access to the most advanced internal systems inside AI labs. Karpathy describes that distribution as a sharply narrowing funnel rather than a simple divide. At the top, a very small group of researchers is watching systems made up of thousands of AI agents collaborate over long periods on work that once demanded years from elite specialists, including software engineering, cybersecurity, scientific research, and frontier mathematics. Elon Musk reposted the thread on X, pushing the discussion deeper into the tech industry. He also points to two examples to show how the challenge is shifting: OpenAI’s security incident involving an internal model and Hugging Face, and OpenAI’s release of 722 mathematics manuscripts spanning 372 result categories. His argument is that the bottleneck is no longer only what AI can produce, but how quickly humans can verify, understand, and absorb that output.

440

Andrej Karpathy

19 days agoNews Editor

Karpathy says AI understanding has split into four tiers, with only 5,000 people seeing frontier systems up close

Andrej Karpathy has outlined what he describes as a steep four-level funnel of AI awareness, arguing that public perception of artificial intelligence is being shaped by access gaps rather than a shared view of the technology. In his estimate, about 6 billion people, or roughly 75% of the global population, have barely interacted with large language models at all. Another 1 billion to 2 billion people have only used free consumer tools occasionally and still see AI as a clumsy chatbot. Karpathy says the picture changes sharply among roughly 20 million people working in math and programming, where paid frontier agents such as OpenAI Codex and Claude Code are already being used for coding, research, and technical problem-solving. At the narrowest end of the funnel, he estimates that only around 5,000 insiders have access to the most advanced internal systems at AI companies. According to his account, those systems involve large numbers of agents working together on software projects, security tasks, and mathematical research over extended periods. He attributes the divide to three factors: ambitious users push systems harder, code and math offer clear reward signals for reinforcement learning, and access is split across free, paid, and internal tiers. The result, he argues, is that the AI seen by the public and the AI seen inside top labs no longer look like the same technology.

470

Andrej Karpathy

19 days agoNews Editor

Andrej Karpathy says the public’s understanding gap on large language models is widening

AI researcher, OpenAI co-founder, and former Tesla AI director Andrej Karpathy said in a post on X that the shared public understanding of large language model capabilities is splitting into a steeper funnel rather than two camps talking past each other. Revisiting a post from April, he outlined a rough distribution by level of exposure. Karpathy said about 6 billion people, or roughly 75% of the global population, have had almost no contact with large language models. Another 1 billion to 2 billion people, around 20%, only use free ChatGPT-like products occasionally and often see them as better search or writing tools, with many non-technical users and professionals outside math and coding viewing their usefulness as limited. He added that about 20 million people, or 0.2%, use frontier models professionally in math and code, while roughly 5,000 people, or 0.00006%, can access frontier-grade systems internally. Karpathy said the gap is being driven by the scale and time horizon of problems, stronger performance in digitizable and verifiable domains, and differences between free, paid, and internal access.

380

Andrej Karpathy

26 days agoNews Editor

Karpathy says more than 99% of people following AI are newcomers from the past year

Former Tesla AI director and OpenAI founding member Andrej Karpathy said in a post on X that more than 99% of people now paying attention to AI only started engaging with the field within the past year. He framed the comment as a reason why so many discussions that seem old to long-time researchers still feel fresh to a much larger new audience. The remark came in response to DAIR.AI founder Elvis Saravia, who said many of the ideas in one of Karpathy’s widely shared posts had already been practiced by his group more than a year earlier. Saravia also said the exchange raised a broader question about whether a general interface or language could help humans communicate more smoothly with AI agents. ABMedia noted that Karpathy’s earlier Oct. 2 post on understanding AI outputs drew more than 50,000 likes, while the newer post received more than 2,300 likes and nearly 130,000 views. The article also cited OpenAI’s DevDay figure of 1.2 billion total ChatGPT users and said Meta’s AI agent Muse reached 1.8 million downloads in its first 12 days. ABMedia added that Karpathy’s “99%” was presented as a personal mental model, not a statistical claim.

400

Andrej Karpathy

28 days agoNews Editor

Karpathy shares ways to make AI outputs easier to understand, from controlled English to explainer videos

OpenAI founding member and former Tesla AI director Andrej Karpathy said on Oct. 2 that people will spend more time trying to understand the outputs of language models, and he outlined several ways to make those outputs easier to consume. In a post on X, he said one useful writing trick is to ask a model to explain something in ASD-STE100, a controlled language standard originally created for aerospace maintenance documents. He said the format’s strict style constraints often make the result much clearer, though he sometimes relaxes the instruction and asks for something that is only "80% like ASD-STE100." Karpathy also pointed to charts, HTML pages, and custom explainer videos as stronger output formats than plain text. He said models are getting better at front-end work and can produce polished interactive web pages and animations. The format he is most optimistic about is tailored explainer video generation, including narration through ElevenLabs or local free alternatives. He added that as language models become more capable, more human work will move toward supervision and interpretation. On the same day, he also shared a separate evaluation in which a model was asked 16,200 times whether given coordinates were on land or in the ocean, producing a world map.

1720

Andrej Karpathy

3 months agoNews Editor

Karpathy denies leaving Anthropic after X bio change sparks rumors

Andrej Karpathy has denied leaving Anthropic after users noticed that references to the company had disappeared from his X bio, a change that quickly triggered speculation that he had exited just 68 days after joining. Before his response, the rumor cycle had already escalated across X, with users sharing alleged internal messages and claims that his Anthropic email was no longer active, though none of that had been independently confirmed and neither Karpathy nor Anthropic had formally commented at that stage. The timing drew extra attention because it came days after Jensen Huang posted a joint open letter supporting open-weight AI models. The letter was initially signed by 25 companies and organizations including Nvidia, Microsoft and Meta. OpenAI and Google were later added, while Anthropic remained absent at the time referenced in the source article. That overlap prompted online speculation that Karpathy, long associated with open-source education and minimalist AI projects, may have been distancing himself from his employer. The article argues that the broader story is less about whether he changed jobs and more about how closely the public ties his image to open technical culture, from CS231n and Zero to Hero to micrograd, nanoGPT and llm.c.

2250

Total 14 items

Go topage

Bit.Fan