Nitin Borwankar

@aakashgupta I made an Apple Silicon MLX version if anyones interested: github.com/trevin-creator…

@wadhwa Not to mention the longest and deepest direct experiential study on what is consciousness. VERY relevant today.

10 years ago I was trying to figure out how to elegantly display a network connection error while kicking off a retry in the background. 10 years later filled with a decade of knowledge and experience…I’m trying to tell Claude how to elegantly display a network connection error while kicking off a retry in the background.

@matt_elevenlabs 11B

We just raised $500M at an 11B valuation 🎉 To celebrate, we’re giving away 1,000 free credits so you can test our platform. For the next 6 hours, comment “11B” below and we’ll DM you the credits (must follow) 👇

For decades, nativists have sold America this narrative that immigrant welfare is behind our deficits and debt. This figure shows how absurd that is. This is all government expenditures, federal, state, and local, for the last 3 decades. Immigrants are not to blame:

@wadhwa No one in the top AI countries, sure. Third world countries with energy scarcity but looking to AI to leapfrog obstacles to progress are probably very concerned about the cost of energy locking them out of the future. Least worst option - get locked into the CN AI ecosystem.

@nitin No one is worrying about the cost right now, the processing is distributed across the internet

China is going to rapidly disrupt the fossil fuel industry by sharply cutting its consumption while driving down the costs of solar energy and batteries, paving the way for other countries to follow suit. This is one of the few good things that it is doing.

@atShruti @taheerBuilds AKA “Safety Second”

@taheerBuilds Everyone's rushing to adopt these tools. Nobody's stopping to think about security.

This is so scary we were attacked 7,922 times over the weekend after using Clawdbot. There are hundreds of Clawdbot servers exposed to the open internet this week. Credential dumps. API keys in plaintext. The thing people missed is that you're not the only input to your agent. Every email it reads, every calendar invite, every webpage it visits is content someone else wrote. Random person DMs you? That's now input to a system with shell access. If you're going to use it, treat it like onboarding a contractor. Dedicated machine or VPS. Separate accounts. Minimal permissions to start. Run it behind Tailscale so it's not exposed to the public internet. Run clawdbot doctor regularly and let it fix itself. The bigger idea is that agents need their own identities. Own devices, own accounts, own credentials. Not piggybacking on yours. Also please stop buying Mac Minis for this. You can run it on AWS free tier in five minutes or a $5 VPS.

@msg Great meeting. Thanks much’

so far we have 3 Claude Code demos scheduled for tonight partiful.com/e/6Hq5PPS75gop…?

my favorite way to use Claude Code to build large features is spec based start with a minimal spec or prompt and ask Claude to interview you using the AskUserQuestionTool then make a new session to execute the spec

@emollick Teach them to articulate their problem well. Teach how to express it crisply and unambiguously in English. Teach them better language skills and critical thinking - these are both surprisingly in short supply. Once they have both these “driving” agentic coding is easy.

Lots of good suggestions in the comments but so far they are aimed at coders (even though they may seem basic to many). Assume that concepts like requirements, unit testing, debugging, etc don’t mean anything to the reader. What aspects need to be taught & what can be skipped?

It would be a good time for experts on coding, and especially experts on programming pedagogy, to think about how to train non-programmers to be good vibe coders. What do they need to know about coding practices in order to be more effective? What limits should they understand?

@RoKhanna @friedberg @chamath How about not taxing unrealized gains but taxing any cash or other realization that uses the unrealized gains as collateral. This keeps the spirit of the law and avoids unintended consequences of taxing unrealized “paper wealth” of entrepreneurs.

So putting aside the rhetorical flourish, you genuinely believe that a 1 percent tax on billionaires wealth for 5 years will kill the SV economy? Honest question. Or your concern is if that is expands to beyond that? Even @chamath has eloquently recognized that those with extreme wealth must do more for society given the backlash and angst people feel.

Peter Thiel is leaving California if we pass a 1% tax on billionaires for 5 years to pay for healthcare for the working class facing steep Medicaid cuts. I echo what FDR said with sarcasm of economic royalists when they threatened to leave, "I will miss them very much."

Announcing New Pareto Frontier on ARC-AGI 27.5% for just $2 333x cheaper than TRM! Beats every non-thinking LLM in existence Cost so low, its literally off the chart Vanilla transformer. No special architectures. Tiny. Trained in 2 hrs. Open source. Thread:

@ElonMuskNews47 Yes

@KeriA1776again Dog thinks “Human feeds me everyday - he must be God.” Cat thinks “Human feeds me everyday - I must be God”

lol

@amooh001 Abundant

Type the first word you see

@ddale8 Have we surpassed 100K Trump Lies yet.

If every random seed is 42 it’s not really a random seed now, is it? IYKYK.

@pamelafox @davetroy I once asked it not to be so optimistic and have a more skeptical and pessimistic attitude to what it was investigating and generating. The next response switched from “Success!” to “Complete Disaster!!” for the exact same situation. 😂

@davetroy Totally agree, I viscerally recoil when I get that affirmation now. You could try prompt engineering, I just use GPT-5 instead. Let me know if you find a stoic prompt for Claude!

I once went to the Stanford "touchy-feely class" where I learnt that I do *not* handle words of affirmation well. 10 years later, and I find I can *not* handle Claude models constantly saying "You're so right!" and "Perfect!" while agentic coding. No words of affirmation, plz!

@itsaj123 @googleaidevs It’s too small for that - you need a 14B or larger model for training on programming languages although there are 3-4B models used for this also. Gemma 270B is good for small targeted tasks like the examples given. You could maybe train it on a *small* macro language or DSL.

@googleaidevs The shown use cases are not that great , the need of hour is , can we train this on lets say frontend programming languages so that we can directly use it on our system with CPU.

Some interesting Gemma 3 270M fine-tuned use cases 🧵⬇️