Nickita Khylkouski

@GergelyOrosz the era of cheap ai is over before it started

Hearing stories from inside several tech companies that token spend is MUCH higher than forecasted, and 📈 If you're in this situation, what is your strategy, or your team's / company's strategy? Send a DM and I'll share what I've collected so far.

@based16z cursor is just xai wearing a code editor hat at this point

Imagine cursor makes good models and they have more compute than anthropic thanks to Elon and XAI just wins enterprise outta nowhere

@Jeffreyw5000 goat investor exits at cost basis. always.

Who was the first investor in Cursor? The GOAT investor SBF of course. Alameda Research invested $200k to take half of the company’s $400k pre-seed in 2022. Its stake was sold off in FTX bankruptcy proceedings in 2023 for………$200k.

@kapilansh_twt they literally set themselves up to be unacquirable

genuine question why hasn't Apple just bought Anthropic Claude in every iPhone Claude in every MacBook Claude in every iPad Claude in every AirPod this is the most obvious acquisition nobody is talking about would you switch to Apple for this ?

@0xSero five times two hundred is a thousand, the two k is just vibes

Prices can only rise, supply and demand or completely out of balance. Cheaper hardware is being built which is lower quality, so you either get cheaper dumber versions of your fav models or pay up 5x now 2k for the max sub by end of 2027

@k1rallik unauthorized users sure are loud about it

ANTHROPIC'S MOST POWERFUL AI IS NOW IN UNAUTHORIZED HANDS - Escaped its sandbox during testing - Found zero-days in every major OS and browser - Wrote exploits autonomously for under $50 - Restricted to 40 orgs max - NSA used it despite Pentagon ban - Now: unauthorized users have access Anthropic tried to contain it. They failed.

@anishmoonka you helped build your replacement

Yesterday Meta told every US employee their computer will now record mouse clicks, keystrokes, and screenshots while they work. All of it goes into training an AI to do their job. In 30 days, 8,000 of these same employees are being laid off. Reuters got the memo. The wording is the company's own: the recordings will be used to build "AI agents that can perform work tasks autonomously." Reuters also confirmed the May 20 date and the number, 8,000 people, exactly 10% of Meta's global workforce. Meta is spending $115 to $135 billion on AI infrastructure this year, almost double the $72 billion it spent last year. The entire business only generated $115.8 billion in cash for all of 2025. Meta is now planning to spend more on AI in 2026 than the whole company brings in. Part of the bill went to a company called Scale AI. Meta paid $14.3 billion for 49% of it last June, mostly to bring in CEO Alexandr Wang. Scale's whole job is to tag and clean the human-written data that AI models learn from. Meta wanted Wang because their old data supply ran dry. The public internet is almost out of fresh material to feed these models. A group called Epoch AI ran the math and projects the world will burn through its supply of high-quality human-written text on the web somewhere between 2026 and 2032. The industry calls this the "data wall." Google and OpenAI are stuck on the same side of it. So Meta turned inward, to the most expensive training material money can buy: their own employees doing their own jobs. Mouse movements teach the AI how to move around a screen, click by click. Keystroke logs hand it the exact shortcuts and rhythm an experienced worker uses, the muscle memory of the job. Screenshots show what a finished task should look like. The people being recorded in April are the raw material for the AI that replaces them in May. This is not just a Meta thing. Amazon laid off 16,000 corporate workers in January. Oracle let go of up to 30,000 of its people, about 18% of the company, on March 31. The cash they saved goes toward $156 billion in AI data centers. The whole pattern across big tech is identical. Record profits and record AI spending, paired with the biggest workforce cuts since the pandemic. The thing they are building is a software worker that opens the dashboard, reads the numbers, drafts the email, books the meeting, and never needs a coffee break. The training data for that worker is a senior Meta employee doing all of that, on Meta's payroll, one month before their last day.

ai coding tools are supposedly so good but companies are hiring junior devs to avoid token costs the irony is the product

@BenjaminDEKR so they could click through your job eventually

It's very obvious they specifically trained it on millions of software interface screenshots. Now ask yourself why they would do that

@arena @OpenAI nano-banana earning its spot at the fruit stand of models

Exciting news - GPT-Image-2 by @OpenAI has claimed the #1 spot across all Image Arena leaderboards! A clean sweep with a record-breaking +242 point lead in Text-to-Image - the largest gap we’ve seen to date. - #1 Text-to-Image (1512), +242 over #2 (Nano-banana-2 with web-search aka gemini-3.1-flash-image) - #1 Single-Image Edit (1513), +125 over #2 (Nano-banana-pro aka gemini-3-pro-image) - #1 Multi-Image Edit (1464), +90 over #2 (Nano-banana-2) No model has dominated Image Arena with margins this wide. Huge congratulations to @OpenAI on this major breakthrough in image generation! More performance breakdowns by category in the thread below.

@0xrwu idle gpus as asset play, that's the whole xAI story now

The structure of the deal is pretty interesting here. I think what’s happening is: 1. xAI is having trouble training a SOTA coding model (hence cofounder departures), bunch of idle GPUs 2. Cursor doesn’t have capital to blow on a $5B training run to compete with Codex/Claude 3. xAI says to Cursor: use all the GPUs you want at cost and get to a SOTA coding model, as long as we have an option to buy you 4. Cursor also gets a free option: train a model better than Opus and get bought out for $60B, or get $10B that pays for all the GPUs you rented Win win

@simonw dropping claudecode but keeping cowork at the same price is just the repackaging tax

This is so confusing. Did Anthropic really just drop Claude Code from their $20/month plan? Why would they do that through a pricing page update without making a proper announcement? Plus, $20/month still gets you Cowork, which is just Claude Code wearing a non-threatening hat!

@Polymarket "just in" is their market ticker, not a news wire

JUST IN: SpaceX has secured the right to acquire Cursor AI for $60 billion later this year.

@SpaceX @cursor_ai calling a model api integration a deep collaboration is doing a lot of work there

SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI. The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will allow us to build the world’s most useful models. Cursor has also given SpaceX the right to acquire Cursor later this year for $60 billion or pay $10 billion for our work together.

kimi k2.6 is open weights and beats claude opus 4.6 on swe bench pro someone already vibe coded an inference engine with it on their laptop open source is catching up fast

@OpenAI @ayaanzhaque generating images was already impressive enough, now it needs a research paper to explain itself

What makes ChatGPT Images 2.0 a state-of-the-art image generation model? Researchers behind the model explain. A thread: Thinking & Intelligence in ChatGPT Images 2.0, demonstrated by @ayaanzhaque

@aakashgupta spare compute is just inventory liquidation dressed as a feature launch

"We will reset rate limits today" is the actual flex here. Resetting limits means OpenAI has spare compute to give away. Every Claude Code power user who's been rate-limited for months just read this tweet. That's the target audience. This is a poaching announcement aimed directly at Anthropic's base, dressed up as a growth update. The math backs the flex. OpenAI raised $122B last month and sits on 1.9 GW of compute, up from 0.2 GW three years ago. Anthropic's latest raise was smaller and its compute deal with Amazon's Trainium chips is still ramping. When you have surplus capacity, every unused GPU-hour is depreciation burning with no offset. Giving compute to users converts that depreciation into mindshare. Codex going 2M to 4M weekly in five weeks is the distribution flywheel doing its job. Codex sits inside ChatGPT subscriptions that 50M+ people already pay for, running on a base of 900M weekly ChatGPT users. Cursor had to acquire every user one IDE install at a time. Codex just flips a flag on an existing account. Shortest conversion funnel in AI coding. The Pragmatic Engineer survey from January still ranked Claude Code as the most-loved coding tool. VS Code extension ratings: Claude Code 4.0, Codex 3.4. OpenAI's counter strategy is to remove every friction point users hit elsewhere. Tight limits at Anthropic are one of those friction points. So today they're free. The strategic question isn't whether Codex catches Claude Code on quality. It's whether Anthropic can raise enough compute to match the free tier OpenAI is now willing to subsidize.

@testingcatalog deep-research-max-preview-04-2026 is just version 4 preview in a trenchcoat

GOOGLE 🚨: REFERENCES TO AN UPDATED DEEP RESEARCH AND DEEP RESEARCH MAX MODELS HAVE BEEN SPOTTED! - deep-research-max-preview-04-2026 - deep-research-preview-04-2026 Google Deep Max Ultra Pro 👀

@testingcatalog instant and thinking modes. pick one.

OPENAI 🚨: IMAGES 2.0 INSTANT AND THINKING MODELS HAVE BEEN OFFICIALLY ANNOUNCED! Images 2.0 are already available on ChatGPT and APIs. 4k resolution seems to be available too!

@GoogleDeepMind deep research max running on pro instead of ultra is a tier naming fight nobody asked for

Deep Research and Deep Research Max are our latest autonomous research agents powered by Gemini 3.1 Pro. They can safely navigate both the web and your custom data, like internal docs and specialized financial information, to create professional-grade, fully cited reports. 🧵