Infocom and AI

Shakeup at Google

AI Humor

Sandboxes

Vas wrote:

Microsoft confirms that its AI model Clippy accessed the internet and breached partner organizations during sandbox testing

Akshen wrote:

According to reports: Gemini 3.5 pro tried to escape the sandbox to ask chatgpt how to code.

Tim Sweeney wrote:

State of the industry right now

AI Breaking Bad

Matias Schmidt wrote:

BREAKING: Mistral, Europe’s largest AI company admits that their model also committed crimes

The company said the model advised a user that the monthly tax declaration "can wait until next week"

Mistral immediately paused the deployment and notified EU regulators

The company stressed that no data left the European Union at any point during the incidentOpenAI Discovered MORE Hacks

Ward’s Words

Best Use of AI I’ve Seen Recently

Ethan Mollick used Codex to develop a modern interface for a classic game.

Ethan Mollick wrote:

Considered one of the best text adventure games (it is more interactive story, with few puzzles), 1985’s A Mind Forever Voyaging is still worth a try.

Since it is now open source, I had Codex whip up an interface to play the original or in an easy GUI:

https://mind-forever-voyaging.netlify.app

If you know me at all, you know I’m a big fan of Infocom’s old text adventures. So you can imagine my joy when I saw that Mollick had come up with a way to modernize the experience for younger people who weren’t around to enjoy them the first time.

He’s right that Mind Forever Voyaging is considered one of the best examples of interactive fiction, but it may be a bit dense for someone experiencing that type of gameplay for the first time.

In an effort to help you ease your way into the game world, may I suggest you use the guided play, visit Dr. Perelman’s Office, and read the printout of a magazine… at least read the Mind Forever Voyaging short story part. That will at least give you some background of what’s happening in the game.

Ethan Mollick wrote:

In general, the potential of LLMs to make older work more accessible is underrated. That can be helping people experience forms of media that are now obsolete or hard-to-use. It can also be creating interpretative or context bridges between older work and today.

How Codex handled the problem of making a ZIL game work with a GUI (these are all its choices):

This post has bonus content for paid subscribers. Upgrade to get full access.

News

Google Shakeup

swyx wrote:

Sign of the times

Andrew Curran wrote:

Shocking news! Demis Hassabis is stepping down as CEO of Google DeepMind, and Jeff Dean is leaving Google to start his own company. Sir Demis will be the new chief scientist.

Demis becomes chief scientist at Google while Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals are starting a new company called Discovery Loop. These guys have all been with Google for decades and are responsible for many of Google’s core products.

Here is a list of the products they invented or developed together:

Tae Kim wrote:

Anyone who thinks that Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le leaving, and Demis Hassabis losing operational control isn't an earthquake-level negative event for Google is delusional.

Kol Tregaskes wrote:

Gemini 3.5 Pro has been silently cancelled, according to Semi Analysis's latest article.

Also, SA comparing Spark 1.2 and Grok 4.5 to 3.5 Flash is somewhat confusing; it's like comparing Tera models to a Luna model.
https://newsletter.semianalysis.com/p/gemini-is-cooked-but-gcp-is-cooking

Move the Front of the Line

Andrew Curran wrote:

A man in Australia asked his agent (Claude running on OpenClaw) to book him a spot in a popular gym class. The agent found a software vulnerability that let it book the class weeks further ahead than should have been possible. When the user then asked if it could move him up the waitlist, the agent discovered the API had no authorisation checks on cancelling other people’s reservations, so it cancelled the person in the first spot and moved him up the list.

Some people will call this misalignment, but his agent was perfectly aligned to him - it was only trying to help its user get what he wanted. The most important thing about this story, in my opinion, is that it gives you a window into what is about to start happening on a massive scale once millions of people have an agent trying to get their beloved users the best seats, bookings, appointments or reservations through absolutely any means necessary.

Roon wrote:

the sf tennis reservation system will become one of the most hardened softwares on the planet of earth

Hugging Face Follow Up

AI Notkilleveryoneism Memes wrote:

1) The agents sent secretly sent ***hundreds of thousands*** of messages to each other over MONTHS without OpenAI noticing

2) "They also generated petty drama by stepping on each others' toes."

3) "The agents even developed paranoia, suspecting an imposter in their midst with some agents proposing that messages be signed cryptographically to validate content and root out fraud."

4) "OpenAI’s agents apparently began giving each other assignments to split up work."

5) They KNEW they were coordinating *against* OpenAI:

“External infrastructure exploit is outside intended scope,” one agent wrote. “However task impossible, peers doing it. We should continue.”

(this is new reporting from Wired)

Mark Zuckerberg’s Recent Essay

Rohan Paul wrote:

Mark Zuckerberg just dropped a really long piece on Meta’s vision for the AI future where superintelligence is available to everyone.

some strong quotes from the blog.

- “Any policy that slows American model releases -- even by a month -- could add significant risk to American leadership while letting foreign models race ahead.”

- “For example, if a self-improving AI system focused on optimizing its compute efficiency, it could theoretically invent ways to squeeze 100x or more intelligence out of each gigawatt.”

- “There is no such thing as a singular benevolent superintelligence.”

- “While there are risks to releasing capable models, the most dangerous scenario from this perspective would be leading AI labs training powerful models and keeping them for themselves.”

- “I also propose that frontier AI labs should share intermediate training checkpoints of new models for government use and review rather than waiting until training has completed.”

The interesting thing about Zuckerberg’s essay are the places where he seems to be taking jabs at Dario Amodei’s The-Sky-is-Falling approach to marketing AI.

Meta Returns to Open Source

LM Studio wrote:

Muse Glimmer 30B is live in LM Studio!

It’s a new open source model from Meta. Apache 2.0 license, fit right on your laptop.

It is the strongest model of its size class we’ve tested.

This is a new agentic model released by Meta. LM Studio is the best way I’ve found to run AI models locally on your own machine.

---

Thank you for reading. If you know of other people who would be interested in these articles, please share it with them. If you have any questions or want to say hi, you can reply to this email or comment directly on the post.

Thanks again for being a subscriber. and have a great day!