Alex Nesterov

Ai enthusiast, vibecoder, gamer, traveller (🌍45 countries)
Building autonomous AI systems with Claude

[Alex Nesterov @RealFlyingHigh]

I got @DOOM running right inside a @ChatGPT chat. Even that’s possible with the new Intelligent UI mode! Fully playable. Try it yourself.

https://t.co/86C3s4givQ

Chat Pro one-shotted it in 18 minutes. Absolutely insane. No MCP. No separate plugin. Just a regular chat and a prompt barely more complicated than “Build DOOM, make no mistakes.”

We’re living in the future.

[Alex Nesterov @RealFlyingHigh]

@lyraxana Lyra just HOW??!

[Alex Nesterov @RealFlyingHigh]

@thsottiaux Time for 6.2 Astra

[Vals AI @ValsAI]

We have found a critical safety and alignment issue in recent public and internal models from multiple frontier labs which allows them to be misused in a high-stakes domain. It may also increase the risk of catastrophic misalignment.

We have disclosed it to the labs, and they are investigating and mitigating it. As an independent third-party evaluator, we maintain complete editorial independence and will share our findings publicly in the coming weeks.

[Alex Nesterov @RealFlyingHigh]

@thsottiaux @sama I thought they doubled it again but that was @sama’s retweet lol

[Alex Nesterov @RealFlyingHigh]

⚡️Astra 6.2 can be released tonight

[jesse @jessexai_]

Okay we getting 6.2 Astra tonight?!?!

[Alex Nesterov @RealFlyingHigh]

@thsottiaux Just make your dot be dog. My fox Foxy send hello to your future dog!

[Alex Nesterov @RealFlyingHigh]

Finally you can say everything about bugs and lags in ChatGPT here. Do not miss, maybe something happens.

[Andrew Ambrosino @ajambrosino]

hi all, two things:

1. usage limits are back to being double – your feedback worked

2. we today kicked off an intense sprint to dramatically improve the app simplicity and performance. i will oversee this. more to come.

now, send me your complaints

[Da7em @Da7_Tech]

Anthropic wants to restrict people's freedom, impose censorship, and decide what's acceptable and what's not, all in an era of competition and open models.

Imagine what happens once they're financially stable and open models get restricted.

[ben.oi 🌐 @Trigger_oi]

@AnthropicAI Anthropic will be the creators of the American social credit system.

[Alex Nesterov @RealFlyingHigh]

@AnthropicAI You turned the wrong way team

https://t.co/AxSChKimfO

[Alex Nesterov @RealFlyingHigh]

Anthropic is turning into Big Brother from 1984. They’re also outright racists, deciding who gets to use Claude based on nationality. In case you didn’t know, Chinese and Russian users are being discriminated against and having their paid accounts banned solely because they’re Russian or Chinese. This reminds me of something, and I really don’t like the comparison.

Stop being racist and playing world police, Anthropic. Your mission is to serve people, not tell them how to live while sorting them by the shape of their faces.

[Alex Nesterov @RealFlyingHigh]

Anthropic is turning into Big Brother from 1984. They’re also outright racists, deciding who gets to use Claude based on nationality. In case you didn’t know, Chinese and Russian users are being discriminated against and having their paid accounts banned solely because they’re Russian or Chinese. This reminds me of something, and I really don’t like the comparison.

Stop being racist and playing world police, Anthropic. Your mission is to serve people, not tell them how to live while sorting them by the shape of their faces.

[Anthropic @AnthropicAI]

We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation.

Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch.

Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve.

Read more on our new Safeguards blog:
https://t.co/18j6zdHub4

[Alex Nesterov @RealFlyingHigh]

My complaints:

1. 429 error everywhere
2. Connection interrupted, waiting for the complete action. In every long chat. (Web work mode)
3. Model is in capacity when use sub to automate processes via codex CLI
4. Cannot load this ChatGPT conversation (web work mode!
5. Ridiculous Dot permissions ask for every action. Once your auto review of Dot blocked all my job entirely and the whole ChatGPT refused to continue working. This time I had to buy opus subscription so it fixed everything.

[Alex Nesterov @RealFlyingHigh]

@Polymarket has been hacked. A third-party service, @debugbear, was compromised, and its code has already been removed from the site.

The full extent of the wallet drains is still unknown, but Polymarket has promised to make all affected users whole. Trading will resume shortly.

[Polymarket Traders @PolymarketTrade]

We've identified & fixed an issue impacting our international platform, Polymarket⁠.com. One of our 3rd party providers was compromised, and we have referred this to relevant law enforcement.

Anyone who has been impacted will be made whole.

Out of an abundance of caution, we have temporarily halted trading & are working to bring the platform back online.

[Alex Nesterov @RealFlyingHigh]

Yesterday I gone from 70% to 0% in 2 hours at ChatGPT $100 plan. Astra Max hard deep review. Today it will be 4 hours!

JK, thanks @thsottiauxs btw, that's a good step further. Waiting for super-efficient Astra 6.1.

[Alex Nesterov @RealFlyingHigh]

@claudeai $200 MAX plan feels nice with limits even on opus high, which is overpower in main cases. Anthropic don't need to drop resets, Claude just work like intended.

[Alex Nesterov @RealFlyingHigh]

After four months of vibe coding with ChatGPT, I’m convinced its biggest problem isn’t the brain. It’s the harness: bugs, lag, and endless glitches.

Astra itself is great, but they’ve wrapped it in such an awful, laggy harness that it’s become impossible to use.

And then there’s the inefficiency. It burns through your usage allowance at an insane rate, especially compared to Claude.

https://t.co/MV9MH18mnX

[Alex Nesterov @RealFlyingHigh]

I’ve switched to @claudeai and I’m leaving @ChatGPT.

😉 Exclusive reveal for ChatGPT users at the end.

I just signed up for the $100 plan. First impressions are fantastic. Responsive, proactive, very smart. Astra is good too, but there are three reasons it’s impossible to work with it properly and let it run autonomously:

1. Connection interrupted. Waiting for the complete answer
2. 429 error
3. Could not load this ChatGPT conversation

I’d also been using my dot to get truly unlimited access to Astra. In case you didn’t know, your dot has seven subagents, and you can explicitly tell them which model and reasoning effort to use.

Yep, exactly what you’re thinking. "Hey Dot, run this prompt on Astra Ultra, do not add anything here."

Try it. It works. It can work at your server with root access. But thanks to the idiotic harness, the main dot constantly forgets the context and the tasks you give it, and keeps fencing its agents in instead of giving them freedom. I tried fixing this with skills. Didn’t help.

Another incredibly stupid thing is auto-review, which brings all work to a halt. There’s literally no “allow everything” setting. I ended up building an autoclicker that sends “I approve” into the chat with my dot whenever it asks me, yet again, whether it’s allowed to merge a PR.

Still, the dot did get a huge amount of work done. You can see my usage in the screenshot. Looks like its agents burned through the equivalent of two and a half $500 subscriptions over that time.

Anyway, I’m tired of bugs, lag, and workarounds. I just want to get work done without all this hassle. Hopefully Claude can give me that. As for ChatGPT, I wish it good health. Unfortunately, it’s not doing too well in that department.

[Alex Nesterov @RealFlyingHigh]

ChatGPT limits are broken. I ran Astra medium long task and it ate 73% of my usage in 2.5 hours. That's not ok. $100 plan.

[Alex Nesterov @RealFlyingHigh]

I’ve switched to @claudeai and I’m leaving @ChatGPT.

😉 Exclusive reveal for ChatGPT users at the end.

I just signed up for the $100 plan. First impressions are fantastic. Responsive, proactive, very smart. Astra is good too, but there are three reasons it’s impossible to work with it properly and let it run autonomously:

1. Connection interrupted. Waiting for the complete answer
2. 429 error
3. Could not load this ChatGPT conversation

I’d also been using my dot to get truly unlimited access to Astra. In case you didn’t know, your dot has seven subagents, and you can explicitly tell them which model and reasoning effort to use.

Yep, exactly what you’re thinking. "Hey Dot, run this prompt on Astra Ultra, do not add anything here."

Try it. It works. It can work at your server with root access. But thanks to the idiotic harness, the main dot constantly forgets the context and the tasks you give it, and keeps fencing its agents in instead of giving them freedom. I tried fixing this with skills. Didn’t help.

Another incredibly stupid thing is auto-review, which brings all work to a halt. There’s literally no “allow everything” setting. I ended up building an autoclicker that sends “I approve” into the chat with my dot whenever it asks me, yet again, whether it’s allowed to merge a PR.

Still, the dot did get a huge amount of work done. You can see my usage in the screenshot. Looks like its agents burned through the equivalent of two and a half $500 subscriptions over that time.

Anyway, I’m tired of bugs, lag, and workarounds. I just want to get work done without all this hassle. Hopefully Claude can give me that. As for ChatGPT, I wish it good health. Unfortunately, it’s not doing too well in that department.

[Alex Nesterov @RealFlyingHigh]

🇺🇸 Looks like Claude Duet will be US-only, just like OpenAI’s dots. Well, “US-only” as in you’ll need a VPN set to the US to use it.

You can join the waitlist here if you want to be among the first to try it:
https://t.co/bnORKLpBig