Systems engineer. I post the tech thing the launch crowd skips. Often unpopular, usually unfashionable, sometimes right.
Austin, TX
Joined December 2012
- Tweets928
- Following380
- Followers191
- Likes1.4K
AMD paying $8.2B for Fei-Fei Li’s World Labs is a pretty huge bet on “physical AI”
chip companies apparently looked at being chip companies and said nah we need the whole stack lol
apple says 4 mac studios can run a trillion parameter model from one wall outlet
$20k desktop suddenly being pitched as the “cheap” ai option is killing me 😭
$NDAQ up while yields are around 5% and oil is above $100
tech stocks apparently saw the macro environment and said whatever
every boring task is a potential SaaS idea
if people complain about doing it repeatedly, there may be money in automating it
7 days autonomous exploring 500+ directions on its own is the part that's actually new here
This is HUGE from @NVIDIA.
Their new AVO agent architecture just scored 100% on the ARC-AGI-3 public set, completing all 183 levels across 25 environments.
The crazy part: Claude Opus 5 scores around 30% as a model baseline. Put it inside AVO, with persistent memory, supervision, tools, and a long-horizon execution loop, and it reaches 100%.
And this isn't some ARC-specific trick. The same architecture previously ran autonomously for 7 days optimizing GPU kernels, explored 500+ directions, and produced kernels that beat FlashAttention-4 by up to 10.5%.
This is a massive demonstration of where AI is heading: the frontier isn't just the model anymore. It's the entire agent system built around it.
agents didn't just speed up coding, they basically doubled the volume of the entire ecosystem
🚨GitHub’s monthly commits just went from 1.4B to 2.9B in four months.
• 130M pull requests merged per month
• 24M new repositories per month
GitHub says the surge in demand contributed to its recent outage.
AI agents are no longer just helping people code faster. They are visibly changing how much software gets made.
ChatGPT Plus is down to $0 right now with a limited-time promo
Here is what you get with it :
Next level models & Deep Research
Codex & Work agents for coding and multi step tasks
Image generation with built in reasoning
Expanded memory & custom GPTs
Check your billing page to see if you can claim it before it expires
Codex is gaining users fast, while Claude seems to be losing them. People are frustrated with Claude’s verbosity, errors, and high usage costs
if this trend continues, Codex could pass 30M weekly users soon, leaving Anthropic playing catch up
TL; DR: Codex is winning, Claude is slipping
prediction markets are pricing under 20% odds astra ships by end of august, and altman's own wording ties the holdup to cyber capability safe guards, not just polish. that's a real delay with a stated reason, not a rumor being debunked
"holding the best model back to snipe the competitor's launch" is a fun narrative but it's still just two x posts and a leak. treat it as a rumor until anthropic actually says something
🚨 Fable 5.1 Update
> A Fable 5 successor, likely Fable 5.1 or 5.5, is reportedly being tested on a subset of Claude accounts
> Some users may already be receiving the model while Fable 5 is selected
> Testing may also be happening inside Claude Code
> Anthropic has used this exact kind of stealth rollout before major launches
> Fable 5 itself was reportedly spotted through limited account testing before its wider release
> The successor has allegedly been ready internally for weeks
> A broader public launch could therefore be very close
Are we about to Get Fable 5.1 Before GPT-6?
a revolver isn't "anthropic needs money," it's the opposite, banks are fighting for a seat at what could be one of the biggest listings ever. that's a confidence signal not a warning sign
Anthropic is reportedly assembling a $10 billion-plus revolving credit facility ahead of its IPO, as Wall Street banks compete for a place in what could become one of the largest tech listings ever.
The important distinction:
Anthropic is not raising another $10B equity round, and it isn't necessarily borrowing $10B immediately.
A revolving credit facility gives the company access to committed bank funding that it can draw, repay, and reuse as needed, providing a large liquidity buffer as its compute and infrastructure spending scales.
And bank demand is reportedly strong enough that commitments could exceed Anthropic's roughly $10B target.
ChatGPT Ads Expanding Across Europe
Next week, ChatGPT Ads are reportedly rolling out to 31 European countries, including Germany, France, Spain, Italy, and the Netherlands
free tier users will now see ads
we're getting dangerously close to a world where frontier models don't get "released" anymore. they just start quietly showing up in your account
the real deepseek moment isn't "open source catches up." it's frontier ai becoming something you can run on your own gpu
frontier ai is turning into a price war. same benchmark score, 5x cheaper output. paying premium prices for marginal gains is getting harder to justify
Grok 4.6 is up to 5× cheaper than GPT-5.6 Sol while delivering the exact same performance on this benchmark
Both score 61 on Artificial analysis
API pricing is insane:
GPT-5.6 Sol:
• $5 / 1M input tokens
• $30 / 1M output tokens
Grok 4.6:
• $2 / 1M input tokens
• $6 / 1M output tokens
That makes Grok 2.5× cheaper on input and 5× cheaper on output
Same performance here....at a fraction of the price
Grok 4.6’s price-to-performance is ridiculous
Cursor launched Origin, a GitHub alternative, just as GitHub went down globally
It brings repos, PRs, code browsing, and AI agents together, with integrations for Vercel, Depot, and Buildkite
Perfect timing to challenge Github
