ivan

co-founder @idler_ai @0xPARC @darkforest_eth @hack_lodge

[ivan @chubivan]

@Jerr_Wu imo should be sooner

[ivan @chubivan]

well, cot censorship is certainly a lever to pull, and it makes sense to pull all available levers given how competitive the token market is. the high order bit on model preference (and therefore revenue) seems to be intelligence as measured by outcome per unit time, closely followed by cost, and everything else (like unavailability of cot) basically rounds down to zero

[ivan @chubivan]

We were surprised to find that despite being the youngest, @grok can run a business better than the rest of the frontier models.

[idler @idler_ai]

We cloned 10 years of sales, inventory, and supplier data from a retail chain into a business simulation. Then, we put frontier AI models in charge for 120 days.

Most made less money than a simple greedy algorithm

Introducing ShelfLife E-Sim: https://t.co/CuEpZLRAVd

[ivan @chubivan]

@0xalpo @cha0sg0d_ @naluconcepcion @FrankieIsLost thank you for your support Alpin! you have made a world of a difference for us

[ivan @chubivan]

I am pleased to finally announce the project that has consumed my life the past couple years: @idler_ai.

Like everyone else, we've been trying to make sense of what in the world is going on with AI. After a year in the RL trenches, we're excited to start sharing some of what we've learned through our work helping to train frontier models.

Besides announcing our company and raise, we're also releasing ShelfLife - a dataset that captures what deploying AI into an ordinary real-world business could look like.

To make ShelfLife, we partnered with a live, profitable, and growing e-commerce store and posed the question: "could AI run this business?". Read the post to find out: https://t.co/xKg0lRfvCw

We'll be releasing more projects like this in the coming months, across coding, law, finance, computer use, and more. Please reach out if you are also pondering what the future holds and want to help shape how all this shakes out.

[idler @idler_ai]

Today, we're introducing idler – a frontier data research lab.

We've raised a $9M seed round, led by @0xalpo and @FrankieIsLost from @paradigm, with participation from @ycombinator, @LongJourneyVC, and angels including @marrc, @catheryn_li, @dcposch, @JanineLeger, @LynettCapital, @feross, and @nur_daulet_.

This funding helps us build the evals, benchmarks, and reinforcement learning environments that the world's leading frontier labs use to train and measure their models.

[ivan @chubivan]

@clkbsfth @idler_ai @0xalpo @FrankieIsLost @paradigm @ycombinator @LongJourneyVC @marrc @catheryn_li @dcposch @JanineLeger Thank you! I am grateful for your small nudge in this direction (:

[ivan @chubivan]

@JackStettner interestingly building a fair rl environment is a fair test for a person

[ivan @chubivan]

the way to solve context switching fatigue isn't to brute force your way through it but to minimize context switching - for instance you can batch similar work into particular days of the week. i interview candidates all day on tuesdays and wednesdays, and do other stuff on other days.

[ivan @chubivan]

@ogandreakiro UGHHH

[ivan @chubivan]

@selkiealt are you striking or stricken?

[ivan @chubivan]

@yacineMTB https://t.co/qx4WjCu737

[ivan @chubivan]

@evanjconrad what's the metric you're measuring?

[ivan @chubivan]

@raidingAI lowk elegant

[ivan @chubivan]

@JackStettner @greptile When’s the next Stettner tweet coming out?

[ivan @chubivan]

@naluconcepcion @JackStettner light mode lowkey classy type shi @DhruvRoongta

[ivan @chubivan]

given this glimpse, i will now be maximizing the rate at which I collect delightful glimpses into lives completely different from mine. this glimpse significantly updated my priors on glimpses: glimpses, like everything, are power-law distributed.

[ivan @chubivan]

@cha0sg0d_ I’m slow you know this