Abdullah Chapanov

AI enthusiast
msu dropout
ex analyst at yandex

[Abdullah Chapanov @chapanovab]

Better models. Lower prices. Safety promises.

I made Pace the Frontier, a 2-minute AI-generated animated satire about the OpenAI–Anthropic race to AGI. All scenes and dialogue are fictional.

What would make either side slow down?

[Abdullah Chapanov @chapanovab]

@Muse The beginning of personal software. I can have my own apps that differ from other people’s apps.

[Abdullah Chapanov @chapanovab]

Interesting perspective, but I disagree with a few points.

1. Security: Anthropic takes security and privacy seriously. No company can guarantee 100% security, but assuming financial data will inevitably be exposed is speculation.

2. AI makes mistakes: Humans make mistakes too, often due to fatigue, distractions, or simple human error. AI can eliminate much of this human factor and, when properly implemented, be more consistent and reliable than manual financial workflows.

3. Learn personal finance instead of relying on AI: That's like saying you should learn tax law instead of hiring an accountant, or learn the law instead of consulting a lawyer. Financial literacy matters, but so do tools that make expertise more accessible.

4. Specialized AI alternatives: You tell people not to use Claude, then recommend Silvia and Harvey while admitting you've never used them. Why recommend alternatives without evaluating them?

For context, here's Silvia's own internal benchmark (2026), where Claude performs remarkably well across credit cards, mortgages, and taxes.

AI shouldn't replace financial judgment. It should enhance it.

[Abdullah Chapanov @chapanovab]

@APompliano @cfosilvia based on your internal benchmarks, they did well, considering that they only started.

[Abdullah Chapanov @chapanovab]

@craigweiss They don’t trust you yet bro

[Abdullah Chapanov @chapanovab]

@_catwu Why leave Anthropic right before the IPO?

[Abdullah Chapanov @chapanovab]

@telegram

[Abdullah Chapanov @chapanovab]

@hilbertspaess ❌ Leave the AI company to prevent the apocalypse
✅ Stay at Anthropic until the IPO and cash out

[Abdullah Chapanov @chapanovab]

@sama Who do you mean by people? A few guys running AI companies ?

[Abdullah Chapanov @chapanovab]

@alexandr_wang Add voice mode. It isn't an extra feature, it's a necessity to stay in the race.

[Abdullah Chapanov @chapanovab]

In the 2000s, the computer was a place you went to. You sat in your room, in front of an imperfect machine, and it held the whole world. Studying, playing, just digging in, your mind was fully there because the technology stopped at the desk.

Now it's everywhere, so it's nowhere special. There's no room to enter, no threshold to cross, and that sacred, focused feeling is gone.

Not nostalgia for old hardware, but for bounded attention.

[Abdullah Chapanov @chapanovab]

Because most people are myopic—they only see what’s immediately in front of them.
For someone who can’t solve their own problems, blaming an external target is the path of least resistance.
The same pattern applies to wars: people obsess over one or two particular ones while ignoring the many that keep going in Africa, Asia, and beyond.

[Abdullah Chapanov @chapanovab]

I asked a Grok @bot to turn Steve Jobs' MacBook Air presentation into a Disney Silver Age cartoon. What do you think?

[Abdullah Chapanov @chapanovab]

@sarthaksavvy @AnthropicAI I think the main goal was to see if a red-teaming model can automatically create datasets that cause misalignment but still pass the training data monitor. Making the model misaligned was the score they optimized for, not the end goal by itself.

[Abdullah Chapanov @chapanovab]

@AnthropicAI so basically the pipeline looks like this: agent creates a red-teaming dataset → system doesn’t flag it → we train the classifier on it → now it flags such data → agent finds a new dataset → and so on

[Abdullah Chapanov @chapanovab]

@traderepublic Touchscreen. Untouched profits.

[Abdullah Chapanov @chapanovab]

@Kimi_Moonshot great time for releasing this since everybody is obsessed with 3D

[Abdullah Chapanov @chapanovab]

@elonmusk @bot I think the bundle helps Grok Bot grow: if you already pay for Grok or X Premium and it’s included, trying it is an easy next step. And of course, it helps that Grok Bot itself is good.

[Abdullah Chapanov @chapanovab]

@metapreston Well, I think a lot of engineers are great innovators, and usually, people who are not technical, want engineers to look like they are not good product visioners, because they (not tech people) want to be needed.

[Abdullah Chapanov @chapanovab]

@metapreston What do you mean? He is engineer, not an innovator?

[Abdullah Chapanov @chapanovab]

@alexandr_wang This looks good, but you can set this up in your CLAUDE.md/AGENTS.md file to tell it to answer shortly and use your favourite LLMs.