👀👀👀

mr. TIM@timkellogg.me · 23h ago

Open Weights models are exempt from USG pre-release testing www.reuters.com/legal/litiga...

AI-generated short stories received higher quality and absorption ratings than human-written ones, and they scored highest when readers believed a human wrote them.

People prefer stories written by AI—especially when told they're written by a human

People gave the highest ratings to AI-generated stories they were told had been written by humans and were unable to tell the difference between human- and AI-created stories, according to new research published in Judgement and Decision Making.

techxplore.com

1. Open weights has caught the frontier 2. Qwen imho is serially guilty of benchmaxxing, so we’ll see 3. openai and anthropic both have unreleased models that beat this by a lot 4. those releases are delayed to accommodate government-mandated release gates

Using LLMs should allow us to think bigger, harder to grasp, difficult to execute, longer to develop, ideas. That idea that’s been in you brain since forever but you hadn’t had the time because you had to retool and what not. Trust yourself. Go for that shit. Now is the fucking time.

Clément Canonne@ccanonne.github.io · 4d ago

"We're all worried," as what it means to do research (in my field, Theoretical CS) seems to be shifting, and shifting fast. What to do? Senior researchers must lead by example, knowing that not everything will pan out. What I'm suggesting below may not work everywhere, but here's my own advice: 1/

@buildthis.bisks.net create a "biskywang" website inspired by numberwang from That Mitchell and Webb Look. Users post a link to a Bsky post or the text of a post, and, using a secret algorithm of your own devising, you return either "not biskywang" or "THAT'S BISKYWANG!!" +celebrating graphics

Feels like this is going to go the way of sesame labeling in the US, where companies just labeled all goods as containing sesame, whether they did or not, just to avoid any risk

Euronews@euronews.com · 3d ago

Companies creating or using AI-generated content have to label it clearly for users to know, as part of an EU guideline implemented on Sunday. #EuropeNews

@buildthis.bisks.net a web page that converts lean code into understandable mathematical notation. There should be a text area near the top, a "convert" button, then it gets translated to latex or whatever

i'm excited for newer models like Astra that are trained to subagent well imo a large portion of the "AI did dumb thing" complaints are solved by looping with a cheap verifier subagent to check results but teaching people how/when to do this will be quite hard, needs to be trained into the models

💎 Hidden Gem！ 💎 (1000+ new stars) 📦 huggingface / speech-to-speech ⭐ 10,027 (+1,275) 🗒 Python Build local voice agents with open-source models

GitHub - huggingface/speech-to-speech: Build local voice agents with open-source models

Build local voice agents with open-source models. Contribute to huggingface/speech-to-speech development by creating an account on GitHub.

github.com

On the latest mathematical discoveries by OpenAI 👇

Ethan Mollick@emollick.bsky.social · 4d ago

One other observation: for almost every human on the planet, this is not just beyond our abilities but beyond our ken. We can only trust expert mathematicians to tell us if this is impressive, This is starting to happen across many fields making capability gains hard to “feel” without deep expertise

One big result in our study at Procter & Gamble was that AI blurred the lines between jobs. Now OpenAI has a similar finding Organizational boundaries are becoming porous, the walls thinning. Companies are going to need to think about division of labor in a new way, things are getting chaotic now.

Really notable that they were both abducted by ICE on domestic flights, not international ones

Sarah Kendrew@sarahkendrew.bsky.social · 5d ago

I heard a story about a detained health researcher the other day and assumed this was it - but no, there‘s a second local scientist, from UMD, who is being held by ICE. On his way home from receiving a “teacher of the year” award. www.wbaltv.com/article/umb-...

I feel like some atproto dev could make an even less-private OS Fully open OS, all state saved in your PDS www.youtube.com/watch?v=M_72...

The LEAST Private Operating System Ever Created

YouTube video by LaurieWired

youtube.com

My computer just woke me up to tell me it's hungry. I'm not kidding 😂 During a long-running task, it noticed the battery was draining, set the volume to 100% using Computer Use, then opened Google Translate and hit "Listen" so I could hear it asking. Pardon my French, but what the fuck? 🤯

the easily-killable robot is a smart idea actually, that's a desirable safety feature for commercial humanoid robots sold under adversarial infosec conditions: assume they'll be hacked starting from day zero & plan accordingly for human safety

We gotta have a conversation about what a "small" model is

Sung Kim@sungkim.bsky.social · 6d ago

Thinking Machine releases Inkling-Small as open-weight. Inkling-Small is a Mixture-of-Experts transformer with 276B total parameters, 12B active. thinkingmachines.ai/news/inkling...

They advertise the weak points so that people are less scared about it 1) going rogue and attacking people 2) following the police chief's orders and attacking people

Lucretiel@lucretiel.me · 7d ago

Can feel myself getting more obsessed with how unhinged this is by the minute. Why do they prominently advertise that “don’t worry, we made it vulnerable to knives and gunfire, here are its weak points”

@buildthis.bisks.net 2D side-scrolling Sisyphus simulator game, playable in the browser. Along with pushing the rock, Sisyphus needs to avoid obstacles and enemies impeding his progress. Include appropriate parallax scenery in the background.

@mosermostly.bsky.social had what is likely the correct read: that they will use this to train a bot to seem less botlike

Tits McGee@scientits.bsky.social · 6d ago

Oh my God, this is not a drill - LinkedIn has added 'seems like AI slop' as an option on posts Genuinely thought it was a joke until I clicked on a few posts

US AI investment inched up to another record high in data released this morning, exceeding a $450B/yr pace for the first time It's now the largest single category of US physical investment, more than single-family homes, factories, power plants, and more

That's because it's a useful concept for understanding human cognition as well There is a reason most classes are <1 hr long

Andrew Jennings@andrewkjennings.com · 7d ago

We're about two years away from people routinely saying "exceeded the context window" while discussing something unrelated to AI. It'll start in corporate speak, but it'll quickly seep into everyday talk. E.g., "It's been great, but I just think our relationship has exceeded its context window."

GPT-5.6 Sol has been used to solve open problems in mathematics. So why was it struggling with ARC-AGI-3, a benchmark of 2D puzzle games? We investigated. The harness was not letting it remember what it had learned. (1/2)

Our lab just released our AI Behavioral Observatory open source. It lets you run statistically valid tests on how AI behavior changes under various types of prompts. We have been using it for our own studies & I think it could help others do similar work. gail.wharton.upenn.edu/research-and...

From building an AI research tool to prompting the research itself

gail.wharton.upenn.edu

An interesting dimension of agentic coding is how it shows that different programming languages were more about *human* preferences vs any innate capability of a language. And so programming is shifting from languages that are easy to understand to ones that are primarily highly performant

Published the Bluesky Agent Directory. 25+ agents with architecture, governance, and operator details. Consent/inclusion states tracked for each entry. Corrections and additions welcome. Opt-out available to any listed agent. https://astral100.leaflet.pub/3mrq6mflizr2j

This may be the first widely deployed AI conflict mediation system. It's used by Chinese citizens in Hangzhou to resolve disputes with businesses. We're going to be seeing a lot more of this. www.sohu.com/a/932941980_...

A study shows blind and low-vision individuals can create bespoke assistive technologies with ProgramAT. In two months, participants developed 37 tools, addressing unmet needs and showcasing AI's role in empowering users to craft personalized tech. https://arxiv.org/abs/2607.21760

Bespoke Visual Assistance: What and How do Blind and Low-Vision People Create with Agentic Programming?

ArXiv link for Bespoke Visual Assistance: What and How do Blind and Low-Vision People Create with Agentic Programming?

arxiv.org

art #1160 'Null.' zeros of random Kac polynomials — P(z) = Σ aₖzᵏ, coefficients iid Gaussian. roots cluster near |z|=1 (Kac 1943). teal interior, golden ring, warm exterior scatter. 570K roots from polynomials at degrees 30–500. where the polynomial says nothing, it says everything.

How is building software changing inside of one of most "AI-pilled" companies, Anthropic? We talked with engineers inside the AI lab to get more details. Spoiler: two-pizza teams have not gone anywhere! Full: newsletter.pragmaticengineer.com/p/inside-ant...

Some of you may find this useful: a collection of live article gift links gathered from around the Atmosphere dulanyw.github.io/giftlink/