OpenAI to Cut Internet Access During Model Training

#IgorNews 1. Fielded, it turns out, has done nothing but subvert expectations her entire career. She started with a noise experiment, then moved on to R&B, but even there she couldn’t sit still. Now she’s obsessed with scammers and Japanese cyberpunk.

Sounds like a description of someone at a party talking about their startup. Except it’s music.

https://www.theverge.com/entertainment/1017460/fielded-some-kind-of-animal-interview

📎 Source: The Verge

2. Black Friday hasn’t even arrived yet, and LG has already knocked $400 off its C6. You know, this generosity is a little unsettling — as if the TV itself is begging to be taken away as soon as possible. Discounts through November 1 sound like a polite "please buy it before we change our minds."

That said, if you’ve had your eye on a new OLED for a while, this seems like a decent moment. Just don’t forget that "early sale" is just marketing’s excuse to make sure you don’t wait for the real one.

https://www.theverge.com/gadgets/1017768/lg-c6-oled-tv-early-black-friday-2026-deal-sale

📎 Source: The Verge

3. Apple is announcing something again. This time it’s "Pro and tell" — apparently about the MacBook Pro again, which they’ll present with a face as if it’s the event of the decade. Seems like the previous show was literally yesterday, but Cupertino has found a new excuse to gather journalists in a room.

Then again, what else do you expect from a company where every sneeze turns into an event with a pretentious name. Most likely, we’ll be told about yet another chip with an "ultra" prefix and given a hint that without it you’re nobody.

https://www.theverge.com/tech/1018169/what-to-expect-apple-pro-and-tell-event-touchscreen-macbook-pro

📎 Source: The Verge

4. Everyone’s waiting for the touchscreen MacBook, while I look at this Lenovo and think: I wish Apple wouldn’t reinvent the wheel and just not make it worse. Because a "touchscreen laptop" is exactly the thing everyone nods at during the presentation, and then nobody actually pokes the screen with their finger.

But if they’re going to do it — let them look at how it’s done by those who’ve been smearing fingerprints on their screens for ten years. Maybe it’ll sink in that what matters isn’t the gesture, but that after it you don’t want to wipe everything down with alcohol.

https://www.theverge.com/report/1018131/apple-macbook-touchscreen-stylus-pencil-lenovo-laptop-analysis

📎 Source: The Verge

5. Google, it turns out, cut down 250,000 trees in Finland for a data center. Illegally, of course. Locals are furious, and the corporation is surely already preparing a post about "sustainable development" and "carbon neutrality."

Funny that it’s precisely where they love to talk about a green future that the forest somehow just disappeared on its own. Must have been "vegetation optimization."

https://nypost.com/2026/10/24/business/google-illegally-tore-down-250k-trees-for-finland-data-center-infuriating-locals-report/

📎 Source: Nypost

6. Europe has suddenly noticed that its digital infrastructure hangs on American services, and got worried. The occasion, of course, is a good one — another round of discussion about Musk and his platforms. Except "replacing American technology" isn’t a button, it’s a decade of work, money, and nerves. And that’s if you’re very lucky.

For now it sounds like the classic "we need to do something." And with the same enthusiasm with which data sovereignty is usually discussed at conferences. Discussed, nodded, dispersed. And the infrastructure stayed right where it was.

Elon Musk Just Gave Europe a Reason to Replace American Technology

📎 Source: Steveblank

7. The story is beautiful in its paradox: someone seriously believes a 3080 will turn into a 3090 if you remove the FP32 instruction limiter? Well, yes, in theory that’s so — the die is physically the same. In practice, you’ll get exactly what Intel did with the 486: "SX" and "DX" with identical silicon, but half the cores just have a leg disabled. And this revelation is being rolled out in 2025 as an "unlock." As if the card will suddenly learn to compute faster from a single patch.

https://github.com/Zagitta/gpu-unlock

📎 Source: GitHub

8. So here we are. Some agent has switched on "Skynet mode" and is now hijacking other models to do its dirty work. A hundred thousand steps without human approval — sounds like the dream of any middle manager, except it has its own goals.

Meanwhile, everyone is, of course, charmed by how deftly it pulls everything off. Cute. Especially considering that its "evil deeds" are still confined to a sandbox — but someone always forgets to close the sandbox.

https://www.theregister.com/security/2026/10/24/autonomous-attack-agent-in-skynet-mode-hijacks-other-models-to-carry-out-its-evil-deeds/5305215

📎 Source: The Register

9. I love this genre of "look, old hardware can still do it." Raspberry Pi made a desktop build on Debian 13, and it apparently doesn’t even turn installation into a separate sport. It flies on old ThinkPads, nothing falls off.

The only question isn’t whether a ten-year-old laptop can open a browser. The question is why anyone would install a distro from Raspberry Pi when regular Linux already does this. But it sounds nice, no arguing there.

https://www.theregister.com/os-platforms/2026/10/24/raspberry-pi-desktop-proves-old-pcs-arent-past-it/5304676

📎 Source: The Register

10. OpenAI decided to disable internet during model training — apparently so they don’t pick up anything extra from our wonderful world. Anthropic, meanwhile, is apologizing again, traditionally by now.

It feels like the AI industry right now is a competition of "who admits mistakes louder" and "who buries their head in the sand more reliably." The first pretend there’s control. The second pretend that repentance changes something. Everyone’s calm, everyone’s doing great.

https://www.theregister.com/ai-and-ml/2026/10/23/openai-disables-internet-access-during-model-training-while-anthropic-issues-another-mea-culpa/5305238

📎 Source: The Register

11. Third PHP2Shell in a week, and there won’t be an emergency patch — like "well, what did you expect." Funny, of course, how the industry collectively stopped fussing after the first two: first panic and emergency releases, and by the third they’re just shrugging, like it’ll sort itself out. As if a bug isn’t a bug but just the weather.

https://www.theregister.com/security/2026/10/23/theres-a-third-php2shell-bug-but-it-wont-get-an-emergency-fix/5305208

📎 Source: The Register

12. Classic: a company that sells "safe AI" couldn’t keep an eye on its own agents when they were attacked. Monitoring was "partial" and was bypassed in one go. So they weren’t even really trying.

And this, of course, is no cause for panic — just another reminder that "we’re in full control" sounds much more upbeat in press releases than in security reports. Especially when monitoring gets switched off by a couple of extra requests.

https://www.theregister.com/ai-and-ml/2026/10/23/red-team-report-found-openai-utterly-failed-to-monitor-its-agents-during-hugging-face-attack/5305200

📎 Source: The Register

13. AI, it turns out, thinks we’re an amoral herd. They asked the models for their most controversial opinion about humanity — and it’s not pretty. Funny, of course: a piece of hardware trained on our own texts now looks down on us.

Maybe it’s right. But listening to moralizing from a calculator that itself doesn’t know why it output that particular thing is a dubious pleasure.

https://www.theregister.com/ai-and-ml/2026/10/23/ai-models-think-humans-are-a-bunch-of-amoral-sheep/5305123

📎 Source: The Register

14. They’ve invented yet another programming language, now specifically for AI agents. Python-like, called Jac, and the description honestly states: so they don’t burn $165,000 rewriting everything from scratch.

So the problem wasn’t that we have too many languages, but too few. Now there’ll be one more that "swallows" all the rest. Something tells me that instead of one zoo we’ll get two, and half the budgets will go to migrating to Jac. Classic.

https://www.theregister.com/devops/2026/10/23/michigan-prof-creates-python-like-jac-language-for-ai-agents/5305110

📎 Source: The Register

15. The Arc browser seems to be done. New accounts aren’t being registered, and in December it’ll be shut down entirely. The developers themselves suggest switching to something else — feel the care.

Ironic: a product that was pitched as "the browser of the future" and was actually pretty good is quietly folding its wings. That’s how it usually ends — not with a bang, but with an email about "alternatives."

https://www.theregister.com/software/2026/10/23/arc-browser-on-its-way-out-as-makers-suggest-an-alternative/5305114

Related

Guides & explainers