Scott Kelly

@andrewchen Same thing for PowerPoint. HTML files are easier for AI coding tools, and more expressive.

prediction re the end of spreadsheets AI code gen means that anything that is currently modeled as a spreadsheet is better modeled in code. You get all the advantages of software - libraries, open source, AI, all the complexity and expressiveness. think about what spreadsheets actually are: they're business logic that's trapped in a grid. Pricing models, financial forecasts, inventory trackers, marketing attribution - these are all fundamentally *programs* that we've been writing in the worst possible IDE. No version control, no testing, no modularity. Just a fragile web of cell references that breaks when someone inserts a row. The only reason spreadsheets won is that the barrier to writing real software was too high. A finance analyst could learn =VLOOKUP in an afternoon but couldn't learn Python in a month. AI code gen flips that equation completely. Now the same analyst describes what they want in plain English, and gets a real application - with a database, a UI, error handling, the works. The marginal effort to go from "spreadsheet" to "software" just collapsed to near zero. this is a massive unlock. There are ~1 billion spreadsheet users worldwide. Most of them are building janky software without realizing it. When even 10% of those use cases migrate to actual code, you get an explosion of new micro-applications that look nothing like traditional software. Internal tools that used to live in a shared Google Sheet now become real products. The "shadow IT" spreadsheet that runs half the company's operations finally gets proper infrastructure. The interesting second-order effect: the spreadsheet was the great equalizer that let non-technical people build things. AI code gen is the *next* great equalizer, but the ceiling is 100x higher. We're about to see what happens when a billion knowledge workers can build real software.

@voxel5d @a_musingcat @OpenAI O3 is by far the most useful.

@a_musingcat @OpenAI What is the best model to use out right now. I don’t have time for all these bad names like o4-mini and 3o maxi blah blah blah. What is the best agentic and non agentic

W. ⏰

@DaylinBritt it's just you

Is it just me or do the refs always make extremely questionable calls in the Chiefs games. Usually benefitting the Chiefs..

Show me exactly 35 posts.

run. it. back.

Sometimes you just have to brute force some faux nested loops, you know?

@simonw Software that exercises discretion to utilize tools (e.g. text output, file search, memory, etc.) in the performance of tasks.

Let’s see if we can crowdsource a robust definition of “agent” (with respect to AI and LLMs) that fits in a <=280 character tweet Reply to this with your best attempt, then scroll through the replies and fave the ones that makes sense to you

Salvador Perez ELITE pitch framer ❗️ #WelcomeToTheCity

Greatest. Game.

I used to run a bcorp tech co and have served on the board of a few tech nonprofits. My (admittedly anecdotal) view is no institutional structure is itself going to achieve that goal. The good news is these companies, however they are structured, are creatures of US law. If and when things get sufficiently serious, the lever for achieving these goals is through are our actual democratic institutions.

We need stronger mission locks on the biggest A(G)I developers (all of them; being a BCorp is not enough), transparent and independent oversight of them, and broader democratic input on whether / how to pursue A(G)I development in general.

OpenAI’s non-profit governance and profit cap are part of why I joined in 2019. If they are removed, it’s as if they never existed—perhaps worse. A (perceived) mission lock can give a false sense of assurance.

@leverus Oh and name of the wind because of course.

@leverus How about combine his craving for games with books? Cough, LitRPG🤓 Ascend Online, He Who Fights with Monsters, Dungeon Crawler Carl, to name a few. For non-LitRPG, you gotta go book 1 of Red Rising but tell him to skim the first 50 pages.

My younger son (12) isn't much of a reader. He craves Roblox, Youtube Shorts, Fortnite, anything but a book. So I gave him the most brutally addictive novel I could think of, ENDER'S GAME ... and it's working. I think. But I can't think what to give him next. Suggestions?

@headinthebox In testing, we found that there was an unstated but very real output limit around 4k tokens where GPT 4 8k and 32k would just flat timeout. GPT 4o just put a ring on the output limit💍

We are being robbed ... OpenAI has silently removed mention of GPT-4 32K (platform.openai.com/docs/models/gp…) and AFAIK they never actually exposed it. Azure is deprecating GPT-4 32K in September (#gpt-4" target="_blank" rel="nofollow noopener">learn.microsoft.com/en-us/azure/ai…) While GPT-4 Turbo, GPT-4o have 128 input context, they only have 4K output context. The same for Claude (#model-comparison" target="_blank" rel="nofollow noopener">docs.anthropic.com/en/docs/models…) and other models. Or am I the lone voice crying in the wilderness?

Big bots (AGIs) are scary in direct proportion to their ability to build very little bots.

Text messaging is my personal Hydra monster. For every piece of political spam I block, I get five more.

Remember when we used to create threads on Twitter that didn't end with a newsletter promo... 🧵 1/2

I'm old enough to remember when the Justin TV guy who wasn't Justin was CEO of OpenAI.

@eshear Or parents are more neurotic and kids used to roam feral 🤷

I am now leading a hypothesis that charts are based on bad data or bad methodology, but unclear

I think about these two charts a lot. What is going on???

@ESYudkowsky I don't think they think it's a good idea so much as they are minimizing legal liability (e.g. avoid hallucinations around citations).

ChatGPT left, Claude right. It's amazing what people think is a good idea to do with RLHF.