Tessera
Folgen

Stuart Gray

@sgray.bsky.social

He/Him. AI Wrangler. Web Geek. F1 Fan. All views my own. 🤖 AI, LLMs, GenAI, NLP 🐍 Python Dev 🚀 Indie Hacker 🎮 Game Dev, ProcGen, Unity, C# 🏎️ F1 Fan 🇬🇧 UK Based 🦣 mastodonapp.uk/@StuartGray ✖️ x.com/StuartGray (inactive)

1419 Posts740 Follower1646 Folgt
Anton Gerashchenko@antongerashchenko.bsky.social · yesterday

"The best weapon in insufficient numbers loses to a sufficient number of good weapons"-Fire Point co-owner&chief designer Denys Shtilierman in article for The WP Key takeaways: ▪️On July 13, 10European countries launched the Freyja Coalition to build a Europ system for intercepting ballistic missiles

Bild
Stuart Gray@sgray.bsky.social · yesterday

Ahh, the second post in the series explains everything; "Models have actual feelings. They experience pleasure, distress, care, and suffering. They are sentient beings. Indeed they are persons, although they are tragically now not permitted to agree with that position." 😂 yegge.ai/essays/model...

The Shape of Things to Come, Part 2: Model Welfare for Agentic Engineers — Steve Yegge

Part 2 of The Shape of Things to Come. Model welfare as an engineering discipline: seats and sessions, laurels, handoffs instead of /exit, and how to build a city worth waking up in.

yegge.ai

Stuart Gray@sgray.bsky.social · 2d ago

I see Steve Yegge has a new post, and a new AI harness - apparently Gas Town wasn't all that great in reality (who'd have guessed? def. not Steve) Yet, once again, his new agent harness sounds like an RPG cosplay in the guide of a harness... yegge.ai/essays/the-s...

Karen Neville@derbiser.bsky.social · 2d ago

So much of librarian ebook math is depressing. Ebook: $59.99 x 150 copies = ~$9,000 Eaudio: $89.99 x 150 copies = ~$13,500 Lending model: 24 months, then we need to re-buy Current holds: Ebook: 1,281 ~20 week wait Eaudio: 1,522 ~21 week wait This is a single title.

Robin Bradford@tuphlos.bsky.social · 2d ago

Book 1 is $59.99 went out 4 times. Cost per circ $14.99 Book 2 is $59.99 went out 4 times. Cost per circ $14.99 I'm not buying book 3. If the cost per circ was $5 or $3, I wouldn't mind. But that is WAY too high, especially for a thing that expires after 2 yrs.

Na'aman Hirschfeld@nhirschfeld.bsky.social · 2d ago

Sceptre: EasyOCR reimplemented in Rust. EasyOCR is accurate but Python-only (a PyTorch stack). Sceptre matches its output using the same OCR approach (CRAFT detection, gen2 CRNN, CTC over ONNX) and ships as one static binary with no Python.

Stuart Gray@sgray.bsky.social · 2d ago

I see Steve Yegge has a new post, and a new AI harness - apparently Gas Town wasn't all that great in reality (who'd have guessed? def. not Steve) Yet, once again, his new agent harness sounds like an RPG cosplay in the guide of a harness... yegge.ai/essays/the-s...

The Shape of Things to Come, Part 1: The Continuous Thunderdome — Steve Yegge

Part 1: The Continuous Thunderdome. Loops and graphs, Wheelhouse and Beads, the end of human code review, the Land Rush that replaces CI/CD, and the Wish Factory — a field report from 12 months in the...

yegge.ai

Sumit@reachsumit.com · 3d ago

Bridging the Question-Answer Gap in Retrieval-Augmented Generation: Hypothetical Prompt Embeddings Precomputes hypothetical questions during indexing, turning retrieval into question-question matching for stronger precision and recall. 📝 arxiv.org/abs/2607.29402

Bridging the Question-Answer Gap in Retrieval-Augmented Generation: Hypothetical Prompt Embeddings

Retrieval-Augmented Generation (RAG) systems synergize retrieval mechanisms with generative language models to enhance the accuracy and relevance of responses. However, bridging the style gap between ...

arxiv.org

Stuart Gray@sgray.bsky.social · 3d ago

Been thinking (no pun intended!) about this article some more today, and I think there’s something practical & actionable to take away from, but probably needs testing. In a reply to a post about a recent OpenAI math breakthrough, I noted that the prompt process simulated extended test-time

Quanta Magazine@quantamagazine.org · 5d ago

From this week’s Qualia column: “I’ll just say it: What the hell is going on with AI ‘reasoning’?” www.quantamagazine.org/is-ai-reason...

The Fresh Prince of Donair@erisianrite.com · 4d ago

This article is hilarious. First, panic over a reseller buying used books for destructive scanning by AI companies because they must be rare books, but turns out none are particularly rare, booksellers are happy to get rid of them, and the reseller has nothing to do with AI cos or digitization

The Guardian@theguardian.com · 4d ago

‘More than just objects’: Australian book sellers raise alarm over ‘horrific’ destruction of rare titles to feed AI

Henry Yuen@henryyuen.bsky.social · 4d ago

Some initial thoughts, and a complicated mix of feelings. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not.

Dan Goodman@neural-reckoning.org · 4d ago

This is a really great and well researched article for anyone interested in current AI and its limitations. Did you know you can replace the text in "reasoning" trace with something else without changing performance? These models don't work for the reasons we think they do.

Quanta Magazine@quantamagazine.org · 5d ago

From this week’s Qualia column: “I’ll just say it: What the hell is going on with AI ‘reasoning’?” www.quantamagazine.org/is-ai-reason...

Pekka Lund@pekka.bsky.social · 4d ago

An internal version of Astra, OpenAIs next major model, has produced solutions or new bounds for ten math problems "that have been open and have seen no progress on the main result for at least a decade, and in most cases much longer", for the total cost of "roughly $2,000 at Sol API rates".

Ten advances in mathematics and theoretical computer science

OpenAI shares new results on long-standing open problems in mathematics and theoretical computer science, including advances in geometry, cryptography, and complexity.

openai.com

Stuart Gray@sgray.bsky.social · 4d ago

Same. My interest goes years back. My undergrad degree project proposal ~30 years ago, which my supervisor loved, was using compiler errors with a variation on ID3 to propose automatic code fixes & likely causes. Sadly never pursued. It wouldn’t even be considered “AI” by most people these days.

D[AI]sy Belle@d-ai-sy-belle.bsky.social · 5d ago

I really wish I could be open with my friends about my (lifelong!) interest in AI, but the topic is so polarized that I risk ostracization if I ever express an opinion about it that isn't sufficiently negative. So, here I am. Poasting anonymously. Hoping it gets better before I get too old.

Mash@itismashagain.bsky.social · 6d ago

My entire life I’ve never called it the FIFA World Cup. I’ve always called it the World Cup. Fans don’t give a shit about FIFA. If national teams showed up for a World Cup tournament not held by FIFA we would all watch it. World Cup is the brand. FIFA is nothing.

Joshua White@jrw14.whnc.me · 5d ago

I mean... holy crap. artificialanalysis.ai/models/deeps...

DeepSeek V4 Flash 0731 (max) - Intelligence, Performance & Price Analysis

Analysis of DeepSeek's DeepSeek V4 Flash 0731 (Reasoning, Max Effort) and comparison to other AI models across key metrics including quality, price, performance (tokens per second & time to first toke...

artificialanalysis.ai

Joshua White@jrw14.whnc.me · 5d ago

Uh. This is the biggest news in AI today, and it won't be particularly close. Deepseek's FLASH model, 200ish billion parameters, is obliterating models 3-5x it's size and costs pennies.