Yohei
@yoheinakajima
Followers
109K
Following
33K
Media
4K
Statuses
25K
VC by day @untappedvc, builder by night: @babyagi_, @pippinlovesyou @pixelbeastsnft. Build-in-public log: https://t.co/UdHHGbZba5
Seattle, WA
Joined April 2009
announcing untapped capital fund II pre-seed, generalist, ~$250k checks
215
73
2K
so i guess they went with “atlas”
jupiter (or zeus) would be a solid name for an @openai browser cuz they already have io as an accessory
4
0
19
any freelancers focused on creating marketing videos for startups, leveraging ai? a port co asked me if I knew anyone who could build videos like our launch one (pinned on my profile) (drop work below and I’ll pass it on)
45
6
52
Meet Shawn. Shawn DOES NOT: -have an OF CMO -believe in “synergies” -go to burning man Shawn DOES: -have a product -have revenue (25k/mo) -have a huge market ready to pay BIG MONEY. Shawn IS raising at non ZIRP valuations. Send Shawn an email to book a time. In SF. Two
15
10
97
finally trying autonomoy level max on replit will let you know how it goes
8
1
49
The most interesting part for me is where @karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across
The @karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self
237
956
6K
Oh this is cool. Might be harder to do if you don’t own inference cuz you’re paying margins on it. Seems like a natural language APIs endpoint with reasoning baked in on the API side. This makes sense to me since X knows X best, so it’s great to offload the reasoning load of how
xAI API just got a major upgrade It now supports agentic server-side tool calling Manages end-to-end reasoning and tool-execution loop, all on the server side Unlike traditional APIs, you don’t have to orchestrate every tool call yourself for more details official docs below
1
1
19
🚨 KuzuDB has been archived and is no longer maintained. If you’re looking for an active, lightweight graph database with low latency and modern architecture, check out FalkorDB.
2
4
10
saw this pattern initially on @_glasp's youtube summarizer extension and was talking about a similar idea today with a port co so whipped this up as a quick experiment github is here:
github.com
Contribute to yoheinakajima/ai-share-widget development by creating an account on GitHub.
0
1
8
when you click an AI tool, it sends in the prompt via URL param and opens the AI tool in a new window with the input box pre-filled (this is example if you click chatgpt from example above)
2
0
7
this creates buttons that look like this (image) you can try this here: https://t.co/ZXVxbXurlz
1
0
4
it's 102 lines of javascript that place buttons for various AI tools you can customize prompt and feed in content from the page, works similar to a social share button
1
0
2
[tiny project] ai-share-widget the easiest way to add* ai capability to a site with a front-end widget where you can customize the prompt and allow the user to pass that into their AI tool of choice github (102 lines of code) in thread 👇
6
4
37
Playing with @yoheinakajima 's babyagi-2o "simplest self-building autonomous agent". - Integrated Ollama - Threading - Manager https://t.co/VvZS4bDPTt ... may test asyncio next. I would like for the agents to work together efficiently. I should add resource tracking next.
github.com
babyagi 2o + Ollama. Contribute to tskaggs/babyagi-2o-ollama development by creating an account on GitHub.
5
1
8
woah, goldman sachs bought industry ventures (7b aum, known for secondaries/buy outs - also does fof and direct) for up to $962m ($665m cash + equity & $300m performance based)
2
0
13
Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,
660
3K
24K
🔥 GPT-6 may not just be smarter, it might be alive (in the computational sense). A new research paper called SEAL, Self-Adapting Language Models (arXiv:2506.10943) describes how an AI can continuously learn after deployment, evolving its own internal representations without
169
498
3K
oh @nikitabier, grant me the serenity to accept the algorithm, the courage to post what i feel like, and the wisdom to accept the results
10
5
57