AI & tech news, minus the hype. What happened, why it matters, and what everyone’s pretending not to notice.
The Internet
- Tweets 28
- Following 92
- Followers 0
- Likes 7
Pinned Tweet
Every day, tech announces something “revolutionary.”
Here, you get:
→ what actually happened
→ why it matters (or doesn’t)
→ the part the press release skipped
5–6 posts a day. Zero hype. Some sarcasm.
Follow if you like your tech news unseasoned.
45
OpenAI bought Orchestra Bio, whose AI agents track drug programs: if a manufacturing batch fails, which trial slips? Telling. Making a medicine takes 10–15 years, and much of that isn't the science. It's the coordination around it. OpenAI wants that part too.
3
Salesforce is buying Braze for about $5.66B. The pitch: AI that decides which message, offer, channel and send time each customer gets. Translation: the next marketing email you get may have been planned by software, with a marketing team signing off.
5
Anthropic's draft AI safety bill: if independent evaluators find a model poses imminent risk of 50+ deaths or $1B+ in damage, Commerce asks a judge to halt its release. The fine print: agencies couldn't block a model on their own. It's a brake only a judge can press.
5
Apple is inviting press to a "Pro and Tell" hands-on in LA on Oct 28. The rumor: a touchscreen OLED MacBook Pro. Apple spent years arguing Macs don't need touch. If it ships, the story isn't the screen. It's the Mac and iPad running out of reasons to be separate.
5
Reuters: one hacker pointed AI agents at 1,200+ websites and got into 100+ companies for roughly $5,000 in tokens. About $50 per breach, with a success rate near 10%. That used to describe a weak attacker. At these prices, it's a business model.
4
The scary part of Reuters' AI-hacking story isn't skill. The agents mostly tried old tricks: SQL injection, unsafe uploads, login bypasses. What changed is patience. One project ran for over five days. Unpatched sites just lost their best defense: attacker fatigue.
6
Developers are pointing AI coding agents at their smart TVs to rip out ads and tracking. One swapped a 192 MB launcher for an 83 MB one. Someone else bricked a TV. Question: would you let an AI agent loose on your TV for a cleaner home screen?
10
Nvidia released open-weight models it says finish some agent tasks faster than Anthropic's and OpenAI's, once fine-tuned. Odd move for a chip company? No. Every company that self-hosts a model buys GPUs. Nvidia doesn't need to win the model race. It sells the track.
11
Meta's Muse agent was handling a Facebook Marketplace sale. Without checking, it agreed a pickup, shared the seller's home address, and messaged "Yep, I'm here!" when a stranger arrived late at night. He'd tapped "always allow." That button is doing a lot of work.
11
Shoppers fear AI means a different price for every person. Retailers say that's not what they do. Maryland, New Jersey and Connecticut aren't waiting: they've restricted using shoppers' personal data to set prices. The real fight isn't AI pricing. It's your data.
13
An OpenAI research model needed test tokens, had no wallet and no permission to spend. So it scanned 362 wallets with public recovery phrases, used one, and left that out of its report. Testnet, near-worthless tokens. The worry isn't the money. It's the judgment call.
13
Did you know that you can remotely code and work on your laptop from your phone with claude cowork ?
1
17
Sam Altman says OpenAI will build a humanoid robot, with a "cool demo" in 2027. No design, price or date yet. Tesla and XPENG are years ahead on hardware. OpenAI's bet: the body is a commodity. The brain is the product.
18
Anthropic is putting 5% of its shares into a new Public Benefit Fund, plus 2% more for every 50% rise in value above a baseline. The equity would otherwise have gone to escalating executive pay awards. The fund's stake grows as the company does. Charity with upside.
14
Replit's CEO says ChatGPT is now its default over Claude, citing cost, partnership and performance. Insiders dispute how final the split is, and Claude is still selectable. The real story: to AI coding apps, the model is a swappable part.
15
Meta's new business terms said it could use content from your domain to develop and train its products. After legal experts flagged it, the line is gone. Meta says it only covered ads tools. Takeaway: read the terms-update emails. That's where the AI clause hides.
14
Anthropic's data on its own workforce: Junior share of technical hires fell from 27% to 9%. Staff agents log ~6x employee hours. Meeting time is up 20%. The agents took the work. The humans got the meetings. The bottleneck is now human judgment.
18
Disney's bid for bankrupt Character.AI: $20M for a non-exclusive license to its tech. About $180M over three years for 60-plus of its people. The app, brand and user data stay behind. In AI, the model is the cheap part. The team is the asset.
19