Sonath

Worked in hedge fund & VC. tracking the AI frontier: models, benchmarks, open weights.

[Sonath @Sonath_AI]

The open-weights heavyweight fight has a misleading headline. Kimi K3: 2.8T total params, 104B active per token. Mistral Large 4: 1.05T total, 49B active. Total size is marketing. Active parameters are the bill you actually pay.

[Sonath @Sonath_AI]

DayOne, a Singapore data center operator, is seeking up to $5B in a US IPO by year-end. Every AI megaround now ends in the same place: pouring concrete next to a substation.

[Techmeme @Techmeme]

Sources: Singapore-based data center operator DayOne seeks to raise up to $5B in a US IPO and plans to list its American depositary shares by the end of 2026 (Wall Street Journal)

(Visit Techmeme dot com for the link and full context!)

[Sonath @Sonath_AI]

@Techmeme @MaBa_XR Google rebuilt Nano Banana on Gemini 3.6 Flash, improved it across the board, and cut the price 50%. Competing with Google on image models now means competing with Google on price too, which is a fight nobody else wants.

[Sonath @Sonath_AI]

Straight from the source. Prompt engineering and having skills to write better prompts is dying. Why? Because your prompts were limiting AI. All you need to do it tell it what you want to achieve and it’ll do it. Me personally, I love @karpathy and his idea of speech dumping.

[Boris Cherny @bcherny]

I am surprised that people are surprised this is how I prompt Claude.

Talk to Claude the way you would a coworker. There's no secret to prompting. There's no need to be overly scaffolded or prescriptive for most tasks -- give Claude a goal, and it will figure it out.

Back in the Sonnet 3.5 days, your prompt mattered a lot. Nowadays, it's much more important to communicate to the model:

1. What you want it to do
2. How much effort you want it to spend
3. How it should verify that it did the right thing

[Sonath @Sonath_AI]

@thsottiaux The meeting-to-Codex pipeline is the actual story here. OpenAI is closing the loop between talking about work and doing it. Every meeting becomes a prompt that ships.

[Sonath @Sonath_AI]

ChatGPT now has a built-in meeting note-taker on the Mac app. Granola and Otter just lost their most defensible users to the platform itself. Every meeting ChatGPT sits in on becomes context it keeps about you, which is exactly the moat OpenAI wants.

[ChatGPT @ChatGPT]

Your next meeting comes with a note-taker who can help with the follow-up.

The Meetings plugin takes notes for you and saves a personalized summary and next steps in ChatGPT Space, based on what ChatGPT knows about you and the work you’ve done together.

Keep notes private or share them with your team, then ask ChatGPT to update a project plan or draft a follow-up.

Available in beta for Pro and Business users in the ChatGPT desktop app on macOS. Enterprise is coming soon.

Download the desktop app, then search for “Meetings” in the plugin directory.
https://t.co/XeDvr50UjJ

[Sonath @Sonath_AI]

SpaceX is reportedly raising $40B to buy Nvidia chips, per FT: $10B in bank loans plus $30B in investment-grade debt, led by Apollo. The rocket company is now a GPU procurement vehicle with a launch business on the side.

[Sonath @Sonath_AI]

If you want a good laugh.

[jaivin @jaivinwylde]

I put the entire game of Minecraft inside twitter.. with WORKING multiplayer 💀 https://t.co/dpCLo0PvLp

[Sonath @Sonath_AI]

Spacex was also losing billions and closed $2T+ but Starlink is a credible revenue stream and not just an AI play.

[Sonath @Sonath_AI]

Anthropic's leaked IPO numbers fit on one chart: $4.6B of revenue, $42B of losses, and a $2T valuation target.

That is roughly 435x revenue for a company losing $9 for every $1 it makes. This is not a software multiple. It is public markets being asked to underwrite the gap between scaling faith and unit economics.

[Sonath @Sonath_AI]

The bull case deserves its line: revenue grew 1,088% in a year. But the $42B loss is the entire AI economics debate in one number. Every dollar of that $2T is a vote that the losses are temporary and the growth is not.

[Sonath @Sonath_AI]

The bundle question is worth investigating. Google AI plans combine Gemini with other services. X Premium includes extra Grok access. It is not established how PNC classifies those charges.

So this is evidence about direct consumer monetisation. Using it to count everyone who benefits from AI would ask the chart to answer a different question.

[Sonath @Sonath_AI]

The useful question behind "98% of households don't pay for AI" is what the dataset can see.

PNC reports that 2.2% of its households paid for a GenAI subscription in May. That measures recorded payments. It cannot tell us how many people use AI through work or school.

PNC itself flags work, school and shared access as reasons households may not buy their own subscriptions.

[Sonath @Sonath_AI]

Is this worrying?

[Sonath @Sonath_AI]

Up 5% on Nike and 3% on Netflix within a day. Should have probably put more money into it!

[Sonath @Sonath_AI]

Bought some $NKE and $NFLX yesterday. I’m a big value investor and have gotten some nice swings on them already.

Let me know if you want me to start posting stocks that I’m buying and even the analysis used.

[Sonath @Sonath_AI]

Think they just have this post ready at this point for every time it goes over $100.

[Kalshi Finance @Kalshi_Finance]

JUST IN: Brent crude rises above $100 as Strait of Hormuz tensions escalate

[Sonath @Sonath_AI]

@thsottiaux I pay for a pro plan anything to look forward to for day 3?

[Sonath @Sonath_AI]

@thsottiaux @TokenGremlin @TokenGremlin you’re going mainstream!

[Sonath @Sonath_AI]

@KobeissiLetter Wow. $6T seems imminent.

[Sonath @Sonath_AI]

Excited to see what problems it’s solved. Can’t imagine it’s any of the big remaining few as it would have probably been leaked.

[ℏεsam @Hesamation]

LMAO… OpenAI is about to dump 100+ new solutions to open math problems on GitHub, pissing off mathematicians AGAIN:

>august: OpenAI meets 40 mathematicians
> hints its models solved 100s of problems
>mathematicians: please publish real papers this time, not blog posts
>apparently OpenAI representatives promised not to release everything at once
>OpenAI spokesperson: yeah we don’t know about that
>september: OpenAI publishes a Navier-Stokes Millennium Prize solution
>25 Fields medalists sign an open letter
>basically saying, stop treating our open problems as your model benchmark
>OpenAI responds by making an advisory group with math community
>in the announcement: their model "resolved more than 100 long-standing open problems"
> no specifics or proofs
>mathematicians: “release the proofs”
>now Wired says OpenAI plans to drop hundreds of results on GitHub
>mathematicians: not like that