Sign in

Holly Jones

@hollyjones1.bsky.social
92 followers 204 following 10 posts
PostsRepliesMedia
Holly Jones @hollyjones1.bsky.social · 30/08/2026
"Don’t think about chunks think how do I represent a unit of information and its provenance to an LLM. and how do I let LLMs navigate that information architecture.”
011
Reposted by Holly Jones
Flavio Martinelli @flavioh.bsky.social · 10/06/2026
NEW PAPER. Why do larger networks train better? "Because they contain more candidate *sub*networks that can learn the task" → lottery tickets This popular explanation uses an appealing but misleading metaphor🧵 We propose an intuitive alternative grounded in theory: escape dimensions
519953
Reposted by Holly Jones
Ethan Mollick @emollick.bsky.social · 22/05/2026
Its funny how much the whole "strawberry" thing, which turned out to be o1-preview & Reasoners, was dismissed as overhyped at launch when it is clear in retrospect that it was way underhyped. A direct line from models unable to do basic math to solving unresolved math problems in 18 months.
51129
Reposted by Holly Jones
Simon Willison @simonwillison.net · 05/10/2025
If you've been trying to figure out DSPy - the automatic prompt optimization system - this talk by @dbreunig.bsky.social is the clearest explanation I've seen yet, with a very useful real-world case study www.youtube.com/watch?v=I9Zt... My notes here: simonwillison.net/2025/Oct/4/d...
youtube.com
Let the LLM Write the Prompts: An Intro to DSPy in Compound AI Pipelines
YouTube video by Databricks
79914
Reposted by Holly Jones
mr. TIM @timkellogg.me · 14/05/2026
Someone posted a real Monet painting but claimed it was AI generated The comments look like a typical bluesky conversation
Alt text for **image_2.png**:
A compilation of social media comments reacting to an AI-generated image created in the style of a Claude Monet water lily painting. The original post asks users to describe why the AI version is inferior to a real Monet.
Key critiques quoted from the comments include:
 * **On Composition and Depth:** "There’s a certain harshness, no soft blending of colors, no depth, no symbiosis of the elements" and "it also seems flatter than a Monet, got depth into his work; his paintings 'move.'"
 * **On Technical Execution:** "The brushstrokes here aren’t as precise, there is less subtle variation in the shades of color, and the overall effect is 'smudgier' than a real Monet" and "the AI seems to be unable to distinguish plant reflections and submerged plants."
 * **On Artistic Intent:** "Art is a receipt of the spiritual growth of the artist. This is not present in 'press button, make Monet' AI copy pasta" and "not made by a human being and therefore not an expression of the human experience."
 * **On Visual Cohesion:** "The reflection of the tree bleeds into the lilypads with no regard for spatial depth or contrast" and "no sense of the threshold between subject and object, just colors."
 * **General Sentiment:** Descriptions range from "high school level art 101 work" and "borked nonsense" to "nice tech demo" but "not a physical painting created by a well known artist."
6654
Reposted by Holly Jones
JA Westenberg @joanwestenberg.com · 14/02/2026
Most people, given a magic wand, would not wish for the ability to write software. They'd wish for their sofware to work properly without them having to do fuck-all. www.joanwestenberg.com/ai-twitters...
joanwestenberg.com
AI twitter's favourite lie: everyone wants to be a developer
Twitter's latest consensus on inevitability: now that large language models can write code, everyone will become a software developer. People, you see, have problems, and software solves problems, and AI removes the barrier between people and software, therefore everyone will build their own software. It's a syllogism, after a fashion,
1203
Holly Jones @hollyjones1.bsky.social · 31/08/2025
Incredible interview: www.youtube.com/watch?v=ziqs...
youtube.com
Surge CEO & Co-Founder, Edwin Chen: Scaling to $1BN+ in Revenue with NO Funding
YouTube video by 20VC with Harry Stebbings
020
Reposted by Holly Jones
Sung Kim @sungkim.bsky.social · 27/08/2025
3 Seductive Traps in Agent Building by Ara Khan In building AI agents at Cline, they've discovered that the most dangerous ideas aren't the obviously bad ones, they're the seductive ones that sound brilliant in theory but fail in practice. cline.bot/blog/3-seduc...
cline.bot
3 Seductive Traps in Agent Building - Cline Blog
In building AI agents at Cline, we've discovered that the most dangerous ideas aren't the obviously bad ones, they're the seductive ones that sound brilliant in theory but fail in practice.
1131
Reposted by Holly Jones
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 21/08/2025
Microsoft made a useful LLM copilot tool that could summarize text in spreadsheets. They provided clear instructions about how to use it and not to use it. In response, journalists are now mocking them for doing exactly the right thing and showing how to use and not use the tools.
1111120
Reposted by Holly Jones
Gergely Orosz @gergely.pragmaticengineer.com · 20/08/2025
The positive thing about AI tooling going mainstream and setting expectations high (e.g. "anyone can build software with AI") is that a large group will learn what us devs know already: Creating good software is hard and it's hard to explain to outsiders in a way they understand
617625
Reposted by Holly Jones
Sung Kim @sungkim.bsky.social · 11/08/2025
All AI-powered code editors, maybe except Github Copilot.
2327
Reposted by Holly Jones
Thomas Frank @thomasjfrank.com · 01/07/2025
A lot of content is made by extremely online people, who observe the behavior of other extremely online people and assume that's how everyone is. It's very easy to let your worldview be shaped by a small number of people who are very good at getting attention on phones
234382
Reposted by Holly Jones
Nathan Lambert @natolambert.bsky.social · 05/06/2025
Video version of my blog from yesterday: A taxonomy for next-generation reasoning models. Skills, calibration, abstraction, strategy (last two are "planning"). Presenting an improved version later today at AI Engineer World's Fair Watch here: buff.ly/b4cvyFw
youtube.com
Traits of next generation reasoning models
Current AI models are extremely skilled, which was seen as the step change in evaluation scores across the industry in the first half of 2025, but often fail...
0174
Reposted by Holly Jones
Geoffrey Litt @geoffreylitt.com · 22/04/2025
what if you could model a problem in a spreadsheet, and then instantly start applying powerful *filters* to explore your space of options? that's the idea @alexwarth.bsky.social and I have been exploring recently at @inkandswitch.com ... 1/
4618
Holly Jones @hollyjones1.bsky.social · 03/04/2025
“Teach people conceptual models that evolve over generations of builders” This line from @davidhoang.com hit hard He asks: What if we evolved how we conceptualize version control before defaulting to Git? That shift could unlock new ways to build software: www.proofofconcept.pub/p/too-legit-...
proofofconcept.pub
Too legit to git
Issue 238: Why vibe coders won't commit
131
Reposted by Holly Jones
Simon Willison @simonwillison.net · 03/04/2025
This is a really neat Python library providing decorator and docstring and type signature based syntactic sugar over my LLM package - I blogged about here: simonwillison.net/2025/Apr/3/s...
simonwillison.net
smartfunc
Vincent D. Warmerdam built this ingenious wrapper around my [LLM Python library](https://llm.datasette.io/en/stable/python-api.html) which lets you build LLM wrapper functions using a decorator and a ...
2565