Prodan

Yapping at scale @resolveai; created Snowflake’s Startup Program, cofounded @laylodotcom and directed a movie

[Prodan @ProdanStatev]

@RecallFielder @JackieFielder_ @sfgov both emails in your form (clerk and mayor) are incorrect

[Prodan @ProdanStatev]

@thsottiaux please fix the Chronicles bug on macos where you can't share your screen on Zoom/Gmeet if you have Chronicles enable

[Prodan @ProdanStatev]

The Supercuts barber, unprompted, asked me if I’m using Claude through Bedrock. in case you’re wondering how SF is doing

[Prodan @ProdanStatev]

That’s what I thought too and have always had Claude and GPT on the biggest model with max reasoning and max thinking budget. But I’ve definitely seen Claude fail on trivial tasks. Also isn’t Anthropic very disincentivesed to tell people to use less tokens if there wasn’t some real reason

[Prodan @ProdanStatev]

@JustinBleuel @dkundel Deep research seems much worse than GPT 5.5 pro extended. Can you merge them or make deep research better? Specifically for things that don’t have a clear end state like product research or finding the best of X (where the model also has to figure out what best means)

[Prodan @ProdanStatev]

@TobinSouth Please improvise your transcription model, it’s much worse than ChatGPT’s or Aqua’s

[Prodan @ProdanStatev]

@msuiche What’s the setup/harness/prompt you use for the agent?

[Prodan @ProdanStatev]

@Plus3Happiness we need better sunscreen in the US first (or people to know that there is better sunscreen available)

[Prodan @ProdanStatev]

@lessin Wolff is the better pick for Insurance Commissioner
https://t.co/HB7cHcD81A

[Prodan @ProdanStatev]

@AndrewCurran_ You should look into his "remarkably prescient" predictions. He did get some things right but he used the wrong models. Eg he predicted we would get to AGI/LEV by reverse-engineering the human brain and have nanobots scanning from inside the brain

[Prodan @ProdanStatev]

@_sholtodouglas “You asked about A, they answered B, and didn’t mention C, which implies D. and now you’re asking me to help you write your POC proposal but what you should really be doing is asking a follow up question to D” (I’m happy to share my real evals privately)

[Prodan @ProdanStatev]

@_sholtodouglas It’s not negotiations. It’s POCs for software. When you’re on a call with your potential customer, the conversation is led by you and what you ask. But since you can’t ask everything, it’s just as important what they didn’t say. So ideally a model can tell

[Prodan @ProdanStatev]

@UtilityCalc @lst156 Your response was more thoughtful and well written than the article itself

[Prodan @ProdanStatev]

@levelsio Keep your bedroom door open

[Prodan @ProdanStatev]

@jacobin this is a great take, thank you @hollyjeanbuck

[Prodan @ProdanStatev]

I'm telling my llm to read only high quality travel content, like the USDA Foreign Agricultural Service annual reports on the food culture and market of that country

[Prodan @ProdanStatev]

@KirkegaardEmil @jensenjeans since the study results are so dispersed, wouldn't it be more epistemically honest to use preCI rather than pooled means? I mean look at those I²