Max Kan

Tokenomics @SemiAnalysis_

[Max Kan @maxkan]

Seems like a forgone conclusion at this point that the best chips in the world in 2029 will be designed by Anthropic/OpenAI

[Max Kan @maxkan]

Nvm, turns out it’s not that goated

[Max Kan @maxkan]

Damn, Ramp data is kinda goated

[Max Kan @maxkan]

@distributionat Bro really do just be tweeting 24/7

[Max Kan @maxkan]

Checkout the Fable 5.2 results on GDPx!

Overall, I think it's clear why they named it Fable 5.2 and not Fable 5.5 lol

[SemiAnalysis @SemiAnalysis_]

Fable 5.2 debuts at the top of the GDPx leaderboard scoring 16.4%. (1/7)🧵

[Max Kan @maxkan]

So ironic that the market sell off and metr report happened on the same day

[Max Kan @maxkan]

Incredible

"What does building two Stargate Abilenes in one year look like? Let’s talk about the Colossus expansions, where Elon is already trying to buy hundreds of nearby homes to both soothe noise complaints and provide construction worker housing."

[SemiAnalysis @SemiAnalysis_]

Elon Is Doing It Again: SpaceX Will Be The World’s Biggest AI Cloud
Making 13GW in 1.5 Years A Reality
https://t.co/n6gXg3Yz1N

[Max Kan @maxkan]

@danielrupawalla Only some of the domains (e.g. finance and legal) use binary because it fits the domain. Most of the domains use continuous scoring.

And even for the domains that use binary grading, you can see the partial scores on the leaderboard

[Max Kan @maxkan]

@dineshtva @btfl_machines Thanks Dinesh! Was a pleasure working with you and the rest of the beautiful machines team

[Max Kan @maxkan]

Regarding the 0% scores, I'm assuming you're referring to legal and finance? Would recommend switching to the partial scoring view for those pages! (score dropdown at the top of the chart)

Both our finance and legal tasks use binary grading because they're high stakes domains: mess up a single key cell in an Excel workbook and the whole thing is worthless.

However, each task decomposes into lots of different criteria the model still earns partial scores for. It's just that none of the models were able to meet the threshold (e.g. 95%) to earn the binary reward. That being said, partial scores can absolutely still be useful for differentiating between models.

[Max Kan @maxkan]

@Yngkvn @jaeahn24 @sujaihiremath @nicoup Was a pleasure working with you Kevin. Definitely need to run it back sometime

[Max Kan @maxkan]

Was a pleasure working with you Akarsh! You + the rest of the Beautiful Machines team are cracked.

And sincere apologies to the @diodeinc team for not tagging them in the main post. That was my miss in the mad rush of publishing. The PCB layout tasks you guys collabed on are incredible.

[Max Kan @maxkan]

@Coulthur You can see the partial scores by clicking the "metric" dropdown at the top. Still very low across the board.

[Max Kan @maxkan]

@JordanNanos lmaooooo, you were cooking with this one

[Max Kan @maxkan]

@moozayus @AndrewMegalaa @ariadata_ Thanks Mo! Was a pleasure working with you.

[Max Kan @maxkan]

@distributionat Appreciate the words of wisdom Mr. Toucan

[Max Kan @maxkan]

@nratliff9 @AndrewMegalaa @dylan522p Congrats Noah!! Was a pleasure working with you all. Super bullish what you and the team are doing at Aria 🚀

[Max Kan @maxkan]

@_rahulparikh @AndrewMegalaa Rahul + the rest of the Living Context team make the best finance data I've ever seen. No question. Was a pleasure working with you all!

[Max Kan @maxkan]

@bryancsk Was a pleasure working with you Bryan

[Max Kan @maxkan]

@uri_rolls @SemiAnalysis_ You’re an absolute legend Uri. Pleasure working with you and Lior!

[Max Kan @maxkan]

@sujaihiremath Sujai 🐐