Bogdan Chechin

Building @onfly_design. Co-founder @ (design for enterprise & gov). Previously: indie products, 15M users

[Bogdan Chechin @bch1601]

@ivanidze404 @WIRED @aimalysheva @mostik_ai So happy for the guys
My friend as well😅

[Bogdan Chechin @bch1601]

@mostik_ai Congratulations guys! Really strong result for just 4 months

[Bogdan Chechin @bch1601]

four months from starting to this. huge congrats to @aimalysheva and the @mostik_ai team 🥂

[Sasha Malysheva @aimalysheva]

meet @mostik_ai!

what happens when you put 12 PhDs in one room for four months? first place on the ARC-AGI leaderboard, which I can't say much about while the competition is still running. and this, which I can.

everyone's arguing about whether open models will catch up to frontier models. we think it's the wrong question. here's the one we pose: why does a frontier model have to generate your answer at all, when the only thing you need from it is the reasoning?

we do this by enabling models to communicate in latent space. through our protocol, hidden states pass straight from a frontier model into a small one running on your infrastructure -- no text between them, and neither model is fine-tuned. two models from different families, sharing reasoning, both left untouched.

how do we know it works? we tested it on a setup where a 753B model reads the problem, and a 4B edge-class model writes the answer. with this approach, we get results 80% as accurate as the frontier model, but at 20x faster performance.

we're committed to preventing frontier model lock-in and are already partnering with inference providers to accelerate open-weight adoption. we've done this between 15 of us, in four months, 12 PhDs and a Fields medalist, backed by @generalcatalyst

WIRED has the first external account of the company and the work: https://t.co/tP8nItCsDl
full writeup, the setup, and all the numbers: https://t.co/C9NZ5vtV1V

[Bogdan Chechin @bch1601]

scrolling twitter WAY too much can cost you 15,000 euros and it's OFFICIAL POLICY

break their rules by viewing over 1,000,000 posts in a day and you owe 15,000 per million. not for posting, for viewing

no joke, ToS section 5, Liquidated Damages: https://t.co/ThDaIdCHEH

[Bogdan Chechin @bch1601]

just found out docusign charges ~20% APR compounded daily if you pay late
legal company doing legal things

[Bogdan Chechin @bch1601]

6 years ago I lived in a corner of a dorm room
now our client owns skyscrapers in NYC

still feels a bit surreal. a multi-billion company, and kid me was obsessed with skyscrapers long before I ever saw one

250 Broadway is one of our favourite recent projects. more in the video

[Bogdan Chechin @bch1601]

@scaling_shields "the enemy of my enemy is my friend"

kautilya, 4th century BC, first known cold email strategist

[Bogdan Chechin @bch1601]

so much agree, and literally every popular post on X is the proof. every single one. open the replies and it's the same slop over and over. half of it paraphrases the post, the other half is "great insight, so true". everyone got handed the same superpower, and that's what comes out

I'm quoting this exact post because it shows the examples best. I've seen the idea around and got to it myself too, but the examples aren't mine: YouTube, Coursera and the internet made world-class knowledge free, it was supposed to democratize everything, most people still chose TikTok

and the experiment is way older than the internet. we already ran it once. okay, not once. lately it runs every 30-50 years: internet, computers before that, electricity before that. but my favorite round is the oldest one. the 1800s: we taught the whole world to read and write. people expected everything to change. the world didn't end, and everyone holding a pen didn't become a writer. we're running it again, just faster. access was never the thing

what actually changed is scale. every market always had its best player, faster and better than everyone. and somehow the mediocre still got their share. the reason is structural: scale has a cost. past some size the best team's processes crack, and client n+1 gets a better result from a worse company that simply has room for them. that's what kept markets split. AI raises that ceiling, hard. routine work, which is most work, now scales

to be fair, this isn't "everything changes now." as we established, everything changes on schedule, every 30-50 years, and we're never objective mid-revolution. it's just one more fundamental shift that redraws the layout of power and money flows. they wrote "winner takes all" in the 90s and it held up fine. soon it might start sounding like they were being careful with words

so the asymmetry grows in both directions: more leverage than ever if you take real responsibility for a piece of the world, less room than ever in the middle. as with every revolution before this one, the middle gets the demotion letter first

and yes, we're in a bubble, an information one. everything changes daily in here, and outside nothing changed at all. most markets haven't even heard the revolution started. so if you can't keep up with the model of the week, take the same leverage somewhere boring. go consolidate plumbing. you'll be early for real, for once, and the edge comes with you. most people won't use it in any market anyway

[Yuchen Jin @Yuchenj_UW]

AI won’t erase the gap between people.

YouTube, Coursera, and Internet made world-class knowledge free. Most people still chose TikTok.

AI is no different. Everyone gets Claude Code, Codex, and ChatGPT. The best will use them better and compound faster.

AI isn’t an equalizer. It’s a multiplier.

[Bogdan Chechin @bch1601]

@speakjava couldn't agree more. AI multiplies whatever you bring to it. bring nothing, get nothing, just faster

[Bogdan Chechin @bch1601]

@Yuchenj_UW twitter proves this every day. everyone's got the same models, and somehow the most popular thing to do with them is slop replies that paraphrase the post above. access was never the bottleneck

[Bogdan Chechin @bch1601]

honestly this is why I stopped worrying about AI changing everything. everyone has the same model now. one guy gets a 300k post out of it, thousands use it to write slop replies under that same post. the tool didn't create that gap, it just showed it. some people do the thing, most don't, button or no button

[Bogdan Chechin @bch1601]

@growing_daniel every home inside of a datacenter, you mean

[Bogdan Chechin @bch1601]

@nestymee Congratulations 🎉 That’s really inspiring

[Bogdan Chechin @bch1601]

@mikekamo the other half of this is volume. wishful thinking creeps in when the pipeline is thin: two candidates and you start imagining qualities they never showed. a wide funnel lets you compare instead of hope. and it's a great question to ask them too: how big are your hr-funnels?

[Bogdan Chechin @bch1601]

@Yungkeem4987 @Byysid that's my point though. disclosure revealed the preference, whatever it's about. if a label changes behavior, it's part of the product. your ad example proves it

[Bogdan Chechin @bch1601]

Cooking something with @seanaiux, @olegrows and our team
Decided to do a Sunday hackathon instead of touching grass

[Bogdan Chechin @bch1601]

@acooldora The engagement is so strong here
Do you consider to become a feetbloger?

[Bogdan Chechin @bch1601]

@Yungkeem4987 @Byysid watermarks didn't kill it, they just made the vote fair. people saw what they were watching and watched less. that's the whole answer really, they don't like it. if the content was actually that good a label wouldn't hurt it. it lost a fair fight, not a hidden one

[Bogdan Chechin @bch1601]

@kirtandopamine for sure it's possible. the main lesson from those deaths: don't position against LinkedIn, interop with it. new thing as the source of truth, LinkedIn mirrors it via API, users lose nothing by joining. you don't beat switching costs, you remove them. happy to discuss

[Bogdan Chechin @bch1601]

@aimalysheva exactly. it's an inverse leaderboard

[Bogdan Chechin @bch1601]

1/ respectfully, this chart swaps one concept for another. the feeling it gives: AI has outgrown humans. what it actually measures: how agents are built. let me unpack

[a16z @a16z]

Humans are the minority user of AI

Agents burn nearly 5x the tokens people do, up 14x since February

Charts of the Week: https://t.co/7YT2BXvuUu