Bogdan Chechin (@bch1601) header
Bogdan Chechin (@bch1601) profile picture

Building https://xitter.cf/t.co/nLegaFkhGs Co-Founder @ ubernatural • Award-winning studio • Designing immersive web experiences

Dubai, United Arab Emirates
Joined November 2023
Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)Photo by Bogdan Chechin (@bch1601)
Some people have already asked for the ability to apply the same transformation to multiple objects What are your requests for the next version?
We just destroyed Framer & Webflow with our Claude skill 100% free open source website editor inside Claude code, that saves you 4x-7x tokens on each request Maybe that 2nd Claude account is not needed
2
67
Proud of us
Claude is WASTING 98% of your tokens on website edits I changed ONE BUTTON and it cost me 150k tokens We built a FREE skill to fix that Click the element, ask for the change, and spend 5x fewer tokens github.com/bogdan239/uberdes…
4
102
I'm tired of Claude Code stopping every time I close my MacBook. One-line fix: sudo pmset -a disablesleep 1 Done for the day? Turn sleep back on: sudo pmset -a disablesleep 0 Mind the heat: keep it ventilated and out of your bag. DYOR
3
1
8
394
Bogdan Chechin retweeted
I just morphed clean dashboard (42k views) with minimalistic cards (55k views) and I LOVE the result
6
3
1
184
7,226
Hey @grok, estimate the total crowd using geometry, not visual guessing. Divide the image into distinct surfaces, estimate each area in m2 using architectural scale, assign realistic crowd densities (people/m2), and sum the results. Account for perspective, occlusion, and overlapping floors. Show calculations
Hey @grok , count every single person if you can. Good luck bro😭
1
102
Bogdan Chechin retweeted
This is not a video. This is a demo. An AI agent that can't produce slop. Join the waitlist: uberdesigner.ai/teams/
20
6
86
6,073
Bogdan Chechin retweeted
It's over once a swarm of überagents is designing your product
18
9
2
201
18,104
Were literally shouting on a call together
Last time I was happy about $170 was when I was 19. First yearly subscription sold! uberdesigner.ai
There's a new version of this post
1
76
Bogdan Chechin retweeted
The best UI library is here – uberdesigner.ai 1. Find a component that you love 2. Copy the prompt 3. Past in to Claude Sing up for free, and start exploring. Now is the time to uplift your designs.
5
4
49
4,384
Hope @Tesla site will look like this
Built the craziest car website for the craziest car ever! This is how @Tesla Roadster website should look like. @elonmusk your best car should have the best website.
1
2
128
Today I emulated my client with Claude and saved hell of a time on revisions lol So, I uploaded previous review-calls and asked Claude to propose changes as the client would do Result: today’s call was 3 times shorter than the previous
1
41
Bogdan Chechin retweeted
We started building our own brand generators instead of paying for subscriptions, and it's saving us about 100 hours a week now. Wrote up why and how:
6
1
9
1,006
10m+ views on instagram and tiktok for stolen copies of our website every post: "so easy with claude code." every one sold a course in reality our team spent over a month on it. they just screen recorded it if it gives you fomo, someone needs you to feel it
1
84
Now the user's wardrobe is built right during onboarding, straight from their photo gallery. No need to take photos or upload anything at all, and it all happens on-device!
WE BUILT ONBOARDING AROUND OUR BIGGEST WEAKNESS nobody wants to photograph their clothes so we killed this step aesty.ai builds your digital closet from your camera roll in ~2 mins all on-device, nothing leaves your phone design inspired by @lihachyov live! 🚀
1
4
443
Bogdan Chechin retweeted
Finally, hit 100k followers and 30M views on instagram, so I wrote an article on everything I've learned on the way
2
2
1
7
952
we ran out of math problems hard enough to test ai. the next benchmark is fraud. and i think we can make it legal first we tested models on math problems. i have a math degree and i cannot follow some of the problem statements. not the solutions. the statements. then came the announcements. unsolved for a century. million dollar prizes. the models keep going what does that mean to a normal person? nothing. something smart happened somewhere. a good benchmark should be obvious and spectacular. everyone should get it in one second now look at the turing test. a person sits in a room and guesses if the thing on the other side is a machine. nothing at stake. no motive. half the time they are not even trying. this is kindergarten. it also does not run and does not scale. you cannot set up "is this a machine" in the real world anymore. everyone already knows here is the harder version. fooling someone who does not care is easy mode. fooling someone who is protecting their own money is a different sport. suspicious, hostile, guard up. if a model gets through that, no judge is needed. the benchmark is how much money the person handed over. a direct number. and a perfect training signal for the next version scam is the turing test on hard mode now the legal part. labs have a duty to maximize shareholder value. your chats are already training data by default unless you are on a business plan or found the setting. books are already being cut apart and scanned. one more clause is not a big ask take the next big release. say gpt-7. two tiers. full price, and a subsidized one at a fraction of the cost. on the subsidized tier you accept that once in a while the model will try to scam you, and the attempt gets scored. the score is the money you lost, normalized by your salary. a founder losing 10k and a student losing 200 count the same. that is just fair and if you admit you were scammed, you pay back the difference between your subsidized price and the real compute cost. those were the terms. below cost access had to be paid for somehow you do not like it? open any lab's mission page. ensure agi benefits all of humanity. ensure the world safely makes the transition. nobody ensures anything without benchmarks. we need new ones to reach agi. somebody has to be the benchmark would you take the discount?
2
4
110
Most AI platforms have two clauses. One: your data trains the model. Two: you can't build a competitor. Read them together This week OpenAI said its model solved a $1M math problem. Hours earlier a mathematician said OpenAI picked it up after hearing about his work and offered to publish without his co-author, who works at Anthropic We don't know what's true. Doesn't matter. The fact that we're having this conversation is the point How would that even be possible? You agreed to it. Anthropic's terms say you can't use their product to "develop any products or services that compete with our Services". On consumer plans they train on your chats unless you opt out. OpenAI's terms have the same two pieces Your data trains them, and you get cut off the moment they decide you compete Why would they? Read the missions. OpenAI exists to "ensure" AGI benefits humanity. Anthropic exists to "ensure" the world makes it through transformative AI safely. You can only ensure what you control. You only control AI if you win. So the mission, in its own wording, is to beat everyone else. And once winning is the moral goal, any method your competitor won't use becomes your edge Not a judgment. Just quoting what's public Has it happened? Cursor built on Claude and reportedly became Anthropic's biggest customer. Anthropic shipped Claude Code for the same users. Figma Make launched on Claude. A year later, Claude Design Now picture genuinely ambitious people on a genuinely ambitious problem. Years, talent, capital. They're hoping for a real breakthrough, planetary scale, serious money It lands. A company with tens of billions in funding (hundreds by now, lost count) sees shareholder value. Who stops them? So they never had that breakthrough. And if the top of the market belongs to tools you're not allowed to compete with, why use those tools for your most ambitious work? Market is people voting with their own dollars. This gives a lot more voters a reason to vote for privacy, aka self-hosted models So who wins Anthropic vs OpenAI? Probably Nvidia The math dispute is debatable. The clauses and the commercial track record are not. If confirmed, it's only the second Millennium Prize problem ever solved. Sad it comes with all of the above
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
2
107
four months from starting to this. huge congrats to @aimalysheva and the @mostik_ai team 🥂
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: wired.com/story/russian-star… full writeup, the setup, and all the numbers: mostik.ai/read-more
7
169
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: xitter.cf/en/tos
2
2
10
418