Excited to share V2 of @usebroca! Simply connect your Google Ads account and we generate new ads for you using AI. Push ads live with one click. Try for free at

@dwarkesh_sp Isn't AI making new discoveries in material sciences, proteins, etc?

@saurav__codes Exa is my favorite

@jeff_weinstein Just sent you an email

@CandlesSarah @ronaldmannak @jzellis Ok so don’t tune it the same way. OP is not asking for a model that is good at creative writing or acts like a therapist. They are asking specifically for a model that does format conversions.

@CandlesSarah @ronaldmannak @jzellis Yes, that is correct. So we agree that LLMs are deterministic. Which means, to your point earlier, a model sufficiently well trained on format conversions with a temperature set to 0 should be reliable enough. Which I believe was the OPs question.

Yes that’s my point. The underlying LLM is deterministic. You are talking about ChatGPT. That is an app that adds randomness to the LLM’s output to make it probabilistic. But if you wanted determinism from it, you would just use the LLM directly and sample only the top probability.

@Digitalgladiato @balajis @Ike_Saul Oh do fuck off, you pedo. Hide behind an anon account from your mom’s basement and cry to someone else. Blocked. Goodbye.

The probability distribution is the same for the same set of inputs. If you sample only the top probability each time, it will always be the same. That makes it deterministic. The argument Sarah made is that an LLM can’t be trusted to do this job because it’s not deterministic, which is incorrect. So by her own argument it should be able to do it. And I assume a sufficiently well trained model would be able to do it.

@CandlesSarah @ronaldmannak @jzellis LLMs are deterministic. It’s only the chat interface (like ChatGPT) that makes it seem probabilistic because of sampling.

@Digitalgladiato @balajis @Ike_Saul Incredible argument. Let me try it on you. I did a quick search of your profile and see that you’ve never spoken out against child pornography. Therefore you must be benefiting from it! You don’t know me mate, so shut the fuck up.

@Digitalgladiato @balajis @Ike_Saul No I don't hate white people. That would be racist too. If this kid made the same comments about whites or any other race, I would say the same thing. Stop projecting.

@balajis @Ike_Saul Isaac is not saying shut down DOGE. He is merely saying you should not award someone for being a raging racist, which is what you are advocating for. Keep DOGE going but place decent people in it. Shouldn’t be too hard.

I'd say this applies to most roles where you aren't required to think deeply about something or be creative. If your work is simply doing a bunch of tasks in order that don't require critical thinking, an AI agent can probably do most of it.

@philosophrob No deal, find someone who pays more. That's the beauty of capitalism and the free markets

@Shpigford @withoptic Looks cool! Why are you selling, Josh?

@amazedsaint @amasad @Replit Just add to the prompt “but make it better” They won’t know what hit them

This shouldn’t even be a debate. The air in India is toxic. It will shorten your life, create health complications, and destroy your cognitive function. The problem is the prevailing attitude in India is “if I work on this, what’s in it for me” and the answer “it’s for the greater good” is not good enough for most people.

AI reply boys (bots) are getting out of hand on this app. @elonmusk y'all need to do something about this or the quality of conversation degrades rapidly. Perhaps limit replies if you're a new or non-paying account?

@deedydas Deedy why you got so many AI replies on your tweet bro 😅

@Shpigford Claude is your friend here