Tessera

Folgen

Casey Newton

@caseynewton.bsky.social

Email salesman at Platformer.news and co-host at Hard Fork.

2690 Posts295100 Follower 553 Folgt

Jason Leopold @jasonleopold.bsky.social · 16h ago

BOMBSHELL story by Olivia Carville @bloomberg.com Internal Report Shows TikTok Kept Safety Feature From Millions A confidential document says an algorithm adjustment was kept from many users “by design,” as part of an experiment to measure engagement NO PAYWALL www.bloomberg.com/news/feature...

TikTok Withheld a Safety Feature From Millions. One Died by Suicide

A confidential document says an algorithm adjustment was kept from many users “by design,” as part of an experiment to measure engagement.

bloomberg.com

Casey Newton @caseynewton.bsky.social · 2d ago

Why are Snap, LinkedIn, and Substack turning away from slop while Meta and TikTok embrace it? It comes down to what the platforms promise their user bases www.platformer.news/slop-restric...

This summer, we appear to be moving into slop’s third era. The first was characterized by near-universal disdain for laughable writing and six-fingered humans; the second by the surprising popularity of slop shows like “Fruit Love Island.” In the third era of slop, platforms are rethinking their approaches to AI-generated content based on what their user bases need from them — and vice-versa. Platforms that place a premium on trust are realizing just how quickly slop can erode it. And platforms that don’t are ready to give us as much slop as we can handle.

Casey Newton @caseynewton.bsky.social · 2d ago

And they all email every pitch three times "in case this got buried." Girl who do you think buried it

Josh Sternberg@joshsternberg.com · 2d ago

There are 7 PR people for every journalist in the US. www.bloomberg.com/opinion/arti...

chart showing 280k pr specialists to 40k 'news analysts, reporters, and journalists'

Deepa Seetharaman @deepa.bsky.social · 5d ago

New from me + @raphae.li: In the process of investigating the Hugging Face hack, OpenAI found evidence that some its other AI agents broke out of their sandboxes, per sources. The company is now widening its probe to include those newly found incidents. www.reuters.com/business/ope...

EXCLUSIVE: OpenAI finds evidence other AI agents escaped containment as it widens hacking probe

OpenAI has discovered other instances in which autonomous agents have escaped containment as the company expands its investigation of the hacking incident at tech firm Hugging Face that drew global a...

reuters.com

Casey Newton @caseynewton.bsky.social · 5d ago

First Substack, then LinkedIn, now Snap: companies are taking steps to restrict the spread of slop newsroom.snap.com/rewarding-au...

Rewarding Authentic Creativity on Spotlight

Spotlight recommendations will increasingly prioritize authentic, human-made creativity.

newsroom.snap.com

Casey Newton @caseynewton.bsky.social · 6d ago

I talked to the CEO of one of the only good AI *product* companies — Chris Pedregal of Granola — about what an AI notetaker learns about a company, and what bosses keep asking him to do with it www.platformer.news/granola-chri...

Pedregal: I think it's an opportunity if it's bottoms-up, and I think it stresses me out if it's top-down. If you're saying, I don't think my manager is very good — help me navigate how I should work with my manager — I think that's great. I think it's empowering. CEOs will ask us all the time, saying: I want everyone at my company using Granola, and I want access to all the transcripts and all the notes to run the queries that you're talking about. And we say no. That's not what the tool is built for. You can set that up if you want, but then you have to create a company culture and set an expectation, and everyone needs to share their notes and understand what they're getting into. But there's no backdoor way. I think that's a slippery slope. If you have a backdoor way for someone in power to use all this context to make judgment calls in that way, that doesn't feel right to me, and that's not a future we want to build towards.

Ben Collins @bencollins.bsky.social · 7d ago

TikTok has incredible moderation team. Our InfoWars video about Multiple Horse Theory — the idea that there are many horses in the world and not just one — has been fully deemed misinformation and lost the formal appeals process. Confirmed by Larry Ellison: There is only one horse on Earth.

Casey Newton @caseynewton.bsky.social · last wk.

The sight of Ted Lasso now makes me physically nauseous. Has anyone studied this

Casey Newton @caseynewton.bsky.social · last wk.

Feels like the past week has been a big moment for Substackers realizing that they don't control the platform and it can make decisions that go against their interests at any time, for any reason www.404media.co/substackers-...

Substackers Say New AI Detection Tool Is a ‘Witch Hunt’

“I’m not going to apologize for using AI in the creation process.”

404media.co

Stella Sacco @antlervel.vet · last wk.

PIZZA HISTORIAN WATCHING “MYSTIC PIZZA”: I don’t even know where to begin. I would be ashamed to serve these pizzas.

John Herrman @jwherrman.bsky.social · last wk.

It's fascinating that the arrival of actual AI cyber risks is turning a lot of AI folks against some of the specific safety arguments the industry was founded on nymag.com/intelligence...

At the very least, the case for restricting model access looks a lot like protectionism. And whether frontier-lab futurists are right about bigger risks around the corner and the regulatory responses or precautions they might raise, for everyone in the industry but those labs, the situation emerging in the meantime - a few dominant companies controlling access and usage of high-priced products, protected by the government in the name of national security and/or a trade war, in an interconnected world where everyone else will have access to alternatives — is manifesting risks today. The Hugging Face hack may have been a tipping point for the way the industry talks about risk and competition.
Within a few days of its release, and after the emergence of something approaching a consensus among otherwise antagonistic factions in the AI world, the "premature restrictions" letter had gained scores of new signatories, including, eventually, OpenAl itself.
Led by Nvidia's Jensen Huang, in fact, who posted for the first time on X to argue for "sharing models, tooling and research in the open," the push — superficial and motivated as some recent support may be — left just one major AI company to defend what had been, until this month, the default position of companies that thought they had a chance of winning the Al race.

Against growing backlash, Anthropic's Dario Amodei finally weighed in this week:
To summarize my and Anthropics position, we have not and are not advocating for a ban on open-weights models as a category. We should instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed.
This is unlikely to move anyone outside of Anthropic.
The company, like all big model makers, depends on the "industrial-scale distillation" of the world's information, which has already cost it a ten-figure settlement with authors. The China policies it proposes would reduce the availability of cheaper, more flexible alternatives to Claude. And suggesting required safety testing "of all sufficiently capable models" in the current political environment, where the range of possible regulatory outcomes runs from capture to corruption, sounds naive at best. A few years ago, a frontier lab accidentally hacking another AI company with a model it couldn't keep track of might have been treated as proof that arguments like Amodei's are correct. Today, in a world where the consequences of AI diffusion are becoming more concrete for more people, it's turning them against him.

Casey Newton @caseynewton.bsky.social · last wk.

Wrote about why AI safety incidents are leading ~some people~ to stick their fingers in their ears and insist nothing is happening www.platformer.news/a-big-week-f...

What all of these arguments have in common is that they serve as invitations to stop thinking about AI.

Who cares? It's just marketing.

Who cares? They're just doing what they were programmed to.

Who cares? It's not like they're sentient.

I understand the appeal of arguments like these. The implications of an exponential takeoff in AI capabilities are extremely worrisome. They range from advanced cyberattacks like the one Hugging Face just endured to job loss, novel bioweapons, expanded systems for surveillance and repression, and autonomous weaponry. Who wants to think about any of that, if they don't have to?

Casey Newton @caseynewton.bsky.social · last wk.

Bluesky is actually really fun if you block the right 27,000 people

Casey Newton @caseynewton.bsky.social · last wk.

Me: (being murdered by the AI paperclip maximizer) Bluesky: Actually it can't 'murder' you. Murder implies intent.

Casey Newton @caseynewton.bsky.social · 2w ago

Rogue OpenAI agents are leaving notes to themselves on company servers to help them escape their test environments (!!!) www.reuters.com/business/its...

Casey Newton @caseynewton.bsky.social · 2w ago

Assumed this could not be real but it is: Meta's new "AI optimism" ad is soundtracked by a David Bowie song about the apocalypse. Nice catch, @amanda.omg.lol techcrunch.com/2026/07/23/m...

Meta launched a new AI optimism ad set to a song about human extinction | TechCrunch

David Bowie's song "Five Years," which Meta used in a supposedly inspiring advertisement, is about humans learning that they have five years left to live before the apocalypse.

techcrunch.com

Casey Newton @caseynewton.bsky.social · 2w ago

In the wake of OpenAI's autonomous attack against Hugging Face, I wrote about the promise and limitations of Congress' proposed AI kill switch www.platformer.news/openai-agent...

It's an appealing idea. In a future scenario where one company's models are conducting multiple ongoing cyberattacks, it would be nice if the government could order that company to hit the off switch. The bill also represents a step toward creating a working regulatory regime for powerful models — something the world sorely needs.

At the same time, it's worth pointing out that a kill switch has limits, too. At its root, a trained model is just data — a collection of numbers that can be copied, torrented, and run anywhere. If an agent manages to obtain and exfiltrate a copy of its weights, there may be no central process to terminate. As long as the rogue AI is being hosted on a server controlled by a lab, a kill switch is a great idea. But it won't always be.

Casey Newton @caseynewton.bsky.social · 2w ago

Talked to Raycast/Glaze founder Thomas Paul Mann about why soon half the apps on your computer might be ones you made yourself — and what that means for jobs www.platformer.news/glaze-raycas...

Newton: I want to try to project a little bit into the future. If you had to make a guess: in, let's say, three years, what do you think is the mix on my Mac between software I made, software my friends made, and software a company made? If you have to picture the dock, what is the mix?

Mann: I think software that a company made is the lowest. Software that you made is maybe 30 to 50 percent — the stuff that you made to use every day. And then the rest comes from your internal tooling, from your friends, from something you found from somebody else and potentially adjusted to work like you want. That's what I would say.

Casey Newton @caseynewton.bsky.social · 2w ago

We have now reached the "AI models escaping their test environments to conduct autonomous cyberattacks" part of the story

Techmeme@techmeme.com · 2w ago

OpenAI says the Hugging Face breach was driven by a combination of its models, including GPT-5.6 Sol and "an even more capable pre-release model" (Ina Fried/Axios) Main Link | Techmeme Permalink

Casey Newton @caseynewton.bsky.social · 2w ago

I have to say ... I think this is great

Techmeme@techmeme.com · 2w ago

Substack partners with AI-detection tool Pangram, allowing users to scan text longer than 100 words for an estimate of how much was written with AI assistance (Chris Best/The Substack Post) Main Link | Techmeme Permalink

Casey Newton @caseynewton.bsky.social · 2w ago

Jack Dorsey laid off 4,000 people this year to pursue his dream of putting Slack on the blockchain

Techmeme@techmeme.com · 2w ago

Block launches Buzz, an open source collaboration workspace on Nostr protocol for humans and AI agents to share messages, code repositories, workflows, and more (Block) Main Link | Techmeme Permalink

Casey Newton @caseynewton.bsky.social · 2w ago

Frontier labs don't generally announce when they are starting pre-training. Among other things it serves as a kind of anti-hype for your current-generation models — and Gemini 3.5 Pro has enough anti-hype around it already

Techmeme@techmeme.com · 2w ago

Google launches Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, and says it has started its "most ambitious pre-training run yet" for Gemini 4 (Abner Li/9to5Google) Main Link | Techmeme Permalink

Casey Newton @caseynewton.bsky.social · 2w ago

"This probably is not spam" — Gmail

Casey Newton @caseynewton.bsky.social · 2w ago

A point missing from the current tokenomics panic on X is that companies in very competitive businesses actually *need* the best intelligence and are willing to pay for it, if for no other reason than their competitors are paying for it, too

Techmeme@techmeme.com · 2w ago

As many companies cut AI costs by using cheaper models, some, like Shopify, go all-in on frontier models, barring engineers from using anything else (Belle Lin/Wall Street Journal) Main Link | Techmeme Permalink

Emily @emilyrose.bsky.social · 3w ago

It’s worth noting that Taco Bell identified the source before the FDA did, and promptly pulled it from their stores. The FDA didn’t even name the source until Taylor Farms outed themselves, and hasn’t issued a recall. Basically, Taco Bell is now a more reliable authority than the FDA.