AI. Robotics. Math.

Joined April 2014
37 Photos and videos
Rachel Kuznetsov interviewed me on KDNuggets - about AI, and about how to be an entrepreneur
Through the lens of a serial entrepreneur, this article explores how the AI revolution is shifting from infrastructure to the application layer, where the greatest opportunities lie in solving specialized, data-heavy in... f.mtr.cool/yytwxqgkfu
25
AndreiRadulescu-Banu retweeted
14 Mar 2025
This is insane. Meet Blender MCP, Claude AI can now talk directly to Blender. Instantly turn any prompt or 2D image into stunning 3D scenes. Creativity unlocked.šŸ‘‡

207
656
5,117
595,811
AndreiRadulescu-Banu retweeted
Really enjoyed this conversation! would recommend taking a listen
Here's my conversation with the founding team of Cursor, a popular code editor (based on VSCode) that specializes in AI-assisted programming. This is a super technical conversation that is bigger than just about one code editor. It's about the future of programming and, in general, the future of human-AI collaboration. It's here on X in full, and is up on everywhere else. Timestamps: 0:00 - Introduction 0:59 - Code editor basics 3:09 - GitHub Copilot 10:27 - Cursor 16:54 - Cursor Tab 23:08 - Code diff 31:20 - ML details 36:54 - GPT vs Claude 43:28 - Prompt engineering 50:54 - AI agents 1:04:51 - Running code in background 1:09:31 - Debugging 1:14:58 - Dangerous code 1:26:09 - Branching file systems 1:29:20 - Scaling challenges 1:43:32 - Context 1:48:39 - OpenAI o1 2:00:01 - Synthetic data 2:03:48 - RLHF vs RLAIF 2:05:34 - Fields Medal for AI 2:08:17 - Scaling laws 2:17:06 - The future of programming
38
113
1,715
306,232
AndreiRadulescu-Banu retweeted
24 Sep 2024
Come join us at MIT this Saturday: app.agihouse.org/events/Agen…
The Harvard/MIT startup scene is absolutely cracked. @cursor_ai, @cognition_labs, @mercor_ai, @Etched all founded by recent grads/dropouts The energy here feels different: no virtue signaling, just insane heads-down building every weekend 😤
1
6
1,895
AndreiRadulescu-Banu retweeted
12 Sep 2024
Happy to release a couple of our reasoning models today (šŸ“)! At @OpenAI , these new models are becoming a larger contributor to the development of future models. For many of our researchers and engineers, these have replaced a large part of their ChatGPT usage. openai.com/index/learning-to… Instead of blurting out an answer right away, ChatGPT can now think through it first. The best analog is that ChatGPT is evolving from using only System-1 thinking (fast, automatic, intuitive, error-prone) to System-2 thinking (slow, deliberate, conscious, reliable). The allows it to solve things it couldn’t before. From a user experience in ChatGPT today, this is a small step forward. On easy prompts, a user likely won’t notice much of a difference (but you will if you have some gnarly math or coding problems šŸ™‚). But this is an important sign of what’s to come.

55
161
1,720
218,485
AndreiRadulescu-Banu retweeted
19 Sep 2024
Some favorite posts about OpenAI o1, as selected by researchers who worked on the model 🧵
143
703
6,132
1,246,809
AndreiRadulescu-Banu retweeted
2 Aug 2024
we crossed 1m downloads* of @latentspacepod! celebratory recapisode with @fanahova special @chatgptapp voice mode demo with @ethansutin! *not even counting spotify which rehosts the file so we dont get stats but estimate 200k
šŸ†• pod: The Winds of AI Winter! latent.space/p/q2-2024-recap The vibes have shifted... @fanahova and @swyx celebrate 1m downloads and recap the last 3 months in AI! Discussing the Frontier Labs vibe shift between Claude 3.5, Llama 3.1, Apple Intelligence, and the expansion of RAG/Ops into the new war for the LLM OS. Bonus: @ChatGPTapp voice mode demo with @ethansutin!
11
4
52
37,998
youtu.be/KcbTbTxPMLc?si=__1p… Bengio interviewed by string theorist Brian Greene: - Dangers of AI - Mechanisms for safety - Could LLMs do science? Yes, in the future, it could, if it searched an answer space. AlphaGo does that. - Mechanistic theory of consciousness
1
3
57
youtu.be/reMnn6bV_fI?si=gEOa… @NoPriorsPod interview with the makers of Sora, the OpenAI model for video generation. Takeaways: - Scale matters. - No word on how training or inference GPU scale compares to LLMs - But, internally, OpenAI has estimated scaling laws
1
1
79
- E.g. when painting brush touches canvas, it leaves paint mark - At current state, Sora does not have 'artistic' optimizations, over just natural training for image generation - Style, in other words, is emergent, not built in - Model still expensive to run
1
19
- Not clear (my interpretation) if economics makes sense yet for mass deployment - Sora should be seen (the makers say) as GPT1 for video generation - With expectation of a Sora2,3,4 to follow, if scaling infrastructure can be sorted out
2
21
AndreiRadulescu-Banu retweeted
Why did I deepfake myself? To see if conversing with an AI-generated version of myself can lead to self-reflection, new insights into my thought patterns, and deep truths.
206
543
2,600
634,335
AndreiRadulescu-Banu retweeted
Some observations/questions: - Did you know that Gemini traffic is already ~25% of ChatGPT? And Google isn't pushing it through their massive distribution channels yet (Android, Google, GSuite, etc). - Big on X, but Claude usage is still very low. Should Anthropic advertise? - ChatGPT is still the big brand, but usage relatively flat over the last year. Why isn't it growing? Is OpenAI compute limited or demand-limited?
169
110
1,343
424,146
AndreiRadulescu-Banu retweeted
GPT-4 Turbo with Vision is now generally available in the API. Vision requests can now also use JSON mode and function calling. platform.openai.com/docs/mod… Below are some great ways developers are building with vision. Drop yours in a reply 🧵
111
420
2,324
1,822,749
In-Person AI Events in Greater Boston, list maintained by Dan Elton and @PaulBaier docs.google.com/document/d/1…

1
17