Back issues
Erich Grunewald argues you should almost never use AI to write anything substantive, and he means the act of putting words on a page rather than the research around it. His three reasons: the writing process is where the thinking actually happens, AI prose is vague and wrong in ways that are hard to catch, and publishing it unlabelled misleads a reader who assumes the words are yours. He is careful about scope. He says transcription, data analysis, search, brainstorming and feedback on drafts are all fine, and so is line editing when a human accepts or rejects every change. The piece is the clearest statement this pool has carried of the case against the thing our own directing desk is named for.
Erich Grunewald argues you should almost never use AI to write anything substantive, and he means the act of putting words on a page rather than the research around it. His three reasons: the writing process is where the thinking actually happens, AI prose is vague and wrong in ways that are hard to catch, and publishing it unlabelled misleads a reader who assumes the words are yours. He is careful about scope. He says transcription, data analysis, search, brainstorming and feedback on drafts are all fine, and so is line editing when a human accepts or rejects every change. The piece is the clearest statement this pool has carried of the case against the thing our own directing desk is named for.
John Hartnup set out to prove that AI event posters do not have to look like the identikit village-fayre style now filling community Facebook groups. He gave ChatGPT invented event details, found that asking for a different style still returned the default template, and kept pushing until the outputs were genuinely varied. The receipts are in the post: each prompt and each image it produced. He ends on a point the paper cares about, that asking for a flat image gives you something nobody can edit, when the same tools will produce HTML or PDF with real text and real layers.
A piece built around what the author calls the fourth terminal: three agent sessions already running, nobody typing, and the reflex to start a fourth. His claim is that this reflex is the defining mistake of the AI coding era, because AI handed people spare capacity and their first instinct was to fill it with more AI instead of asking what the capacity was for. He is fair to the tools first and describes the honeymoon in detail, work that took a day finishing before lunch, a quarter-old migration drafted in an afternoon. Then he spends the rest of it on what four months of that cost him, with the research cited and not gestured at. It pairs directly with the Grunewald essay elsewhere in tonight's pool.
A guest argument on Terry Tao's blog says mathematics needs to give real academic credit to what it calls motivated explanations, now that proofs can be generated without the understanding they used to stand for. The author writes from outside academia, makes videos for a living, and says so up front because the proposal favours exactly the work he does. His diagnosis is stronger than his pitch: if the field keeps celebrating proof generation above clarification, outsiders will keep concluding that proof-generating machines make mathematicians obsolete. Every field that has a proxy for its real goal is about to have this argument, which is why it belongs on the amplifier desk and not in a maths section.
A blogger reports that GPT-6 Astra decoded a German ADFGVX cipher message from November 1918 that he believes had never been solved, one of about a dozen surviving messages that had resisted earlier efforts. The post shows its work: the encryption word the model chose, how the table is built from it, and the decoded German with an English rendering. The result is a single anecdote from one person and nobody independent has checked it. Read it as a capability sighting rather than a measurement.
The New Stack's week in review, arguing that companies are now buying the harness around inference rather than the inference itself. The supporting numbers: Vercel's AI Gateway reported average price per token fell 23.2% in August, a third consecutive monthly decline, and Zed says 33 of its team landed 570 commits through its new Delta tool. Zed's pitch is that a diff shows where an agent ended up while leaving the conversation that got it there somewhere else, so Delta keeps the thread attached to the code. Worth noting at the foot of the piece: the publisher's owner, Insight Partners, is an investor in Anthropic, one of the companies the roundup covers.
A founder's account of Laya, an open-weight sub-35ms decision engine that returns typed probabilities over schemas instead of generating text, aimed at classification, guardrails, routing and triage. Half the post is a priority grievance: Nandakishor Mukkunnoth says he published this architecture in an arXiv paper in March 2025 with weights and a dataset, and that TypeSafe AI launched Jev in September 2026 on the same non-autoregressive idea without a technical paper. The speed and accuracy figures are the house counting its own shelves, and the benchmark board is his. The claim worth reading is the architectural one, and it lands the same week two other rows in this pool are about Jev.
Latent Space's AI news roundup, and the through-line is how fast TypeSafe AI's Jev is being cloned. Numbers worth keeping: on the RSI-Exam leaderboard GPT-6 Astra holds first at 0.5126 with Fable 5.1 entering second at 0.4813 and no model reaching the frontier-calibrated reference; on the newly launched CUA-Bench for real keyboard and mouse use across six games, every frontier model scores below twenty percent. It also carries a taxonomy dispute about what counts as recursive self-improvement, turning on whether a system can modify its search strategy and research tooling rather than only its own weights. The body stops at a free-trial wall partway through, so the Reddit recap and everything after it did not reach the pool.
Trump rejected calls to slow AI development, promised an AI Force and floated a role for the technology in law enforcement, with the line that we will cherish it. The reporting around the quote is the substantial part. A researcher named Jacob Coxon resigned from Anthropic days earlier accusing both Anthropic and OpenAI of racing to self-improving superintelligence, and Anthropic's alignment science lead Evan Hubinger replied in public that we really do earnestly believe AI could kill all humans. A lawsuit filed the same week alleges the labs coordinated a slowdown in violation of antitrust law. All of this is policy and industry politics with no practice in it, which puts it on the wire and not at a desk.
Dan Abramov's long first-person account of driving AI agents at a proof of Conway's conjecture on surreal numbers, and the piece is at its best where the workflow fails. He describes nudging mathematical agents through claimed breakthroughs until the stack of generated papers grew from a dozen to almost thirty in a day, with the knowledge that one mistake anywhere would invalidate everything after it. Lean verification ran so far behind that it added no assurance at all. He names the failure mode precisely: the model would assert a path existed, then later assert an obstruction, and burn hours proving neither. This is the most honest record of an agentic loop eating its own tail that the pool has carried, written by someone who kept going anyway.
An empirical study of coding-agent harness design that holds the execution loop fixed and varies three components: planning, action space and context management. The authors ran 176 matched settings across four models on SWE-Bench Verified and Terminal-Bench 2.1, covering five context-management strategies and four context-window budgets. Four findings: context management matters more as the window tightens and most of its value is in preventing overflow failures; rule-based elision staged before LLM summarization is the most efficient strategy, while making elided content recoverable adds machinery models rarely use and buys no accuracy; planning acts as an accuracy scaffold for weaker models and a cost saver for stronger ones; predefined tools help models with weak bash skills, while bash-capable models do fine and cost substantially less on a bash-only interface. This is the measured version of arguments the rest of the pool is having from anecdote.
LangChain asking whether TypeSafe AI's Jev, a model that returns typed answers and probabilities instead of generating text, could be a third kind of agent evaluator alongside code-based checks and LLM-as-judge. The framing is useful: code evaluators are cheap and reliable but only work on narrow set-input problems, while LLM judges are slow, expensive and unreliable. The speed and cost figures, up to 200x faster inference and 400x lower cost on classification, are TypeSafe AI's own and the post says so. This is one vendor assessing another vendor's model, with a joint webinar announced in the middle of it, so read the idea and discount the numbers.
A practitioner's walk through entity resolution on a supplier list of just over 10,000 websites, where the same publisher arrives as four spellings and a duplicate record means paying an invoice twice. The argument is that string matching is the easy part and deciding what to do with the scores is the actual job, so two boring deterministic stages beat the similarity model. The pipeline is run end to end on a synthetic rebuild with ground-truth labels, which makes every precision claim in the piece checkable rather than asserted. An honest aside worth keeping: one marketplace they sync tells them its listed prices are right about ninety percent of the time.
The New Stack on Lauren Tan, an engineer on the Grok team at SpaceXAI, whose personal agent workflow has her shipping 2,000 pull requests a month to production, close to a hundred per working day. Her own claim is that verification carries the whole thing: a skill that lets the agent check its work and keep going, which she treats as critical infrastructure and not as one skill among many. The article argues that the skill only works on top of a rich runtime the agent can drive and get structured answers from, which a single application provides for free and a system of hundreds of services does not. Two things travel with that number. It is relayed from Tan's own published guide and nobody has measured it independently, and Signadot, which sells exactly the multi-service runtime the piece concludes you need, sponsored the post.
Four named plaintiffs who pay for ChatGPT, Claude, Grok and Gemini subscriptions are suing Anthropic, OpenAI, SpaceXAI and Google, claiming the companies broke antitrust law by agreeing to slow AI development. The alleged coordination is dated to September 12, when Dario Amodei published an essay urging industry-wide deceleration in favour of safety and Altman, Musk and Hassabis each publicly agreed the same day. The plaintiffs say they are not opposed to the labs seeking regulation or an antitrust exemption through government, only to a private agreement among rivals. A live legal test of whether safety coordination is a cartel, which is worth the paper's readers knowing about even though there is no practice in it.
Google's AI and Infrastructure team describing how it embeds agentic vulnerability scanning into its development lifecycle, scanning every code change across hundreds of millions of lines rather than running large one-off scans. The design argument is the interesting part: pre-submit scanning of one change at a time needs far less context than analysing a whole codebase, and it lands in the tools developers already use, so security behaves like a readability review instead of an event. The headline claim, hundreds of vulnerabilities per month prevented from reaching the codebase or production, is Google measuring its own programme with no external check. The shape of the approach is the transferable part.
Today's paper: 16 stories, 5 on the front page, 6 desks staffed.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside one more today: “AI-generated posters don’t have to be horrible”.
erichgrunewald.substack.com — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Our page is the summary; erichgrunewald.substack.com's is the story. Click through — they earned it.
THE DIRECTING DESK today: 2 stories.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside one more today: “AI-generated posters don’t have to be horrible”.
erichgrunewald.substack.com — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Our page is the summary; erichgrunewald.substack.com's is the story. Click through — they earned it.
THE AMPLIFIER DESK today: 3 stories.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
Made for anyone whose day holds more work than hours, whether they know it yet or not. The move: try it on the one task that ate yesterday. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside 2 more today: “If math is more than proof, we need to better celebrate the rest of it”; “GPT-6 Astra Solves a WWI German Radio Cipher”.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Our page is the summary; Towards Data Science's is the story. Click through — they earned it.
THE LOOP DESK today: 5 stories.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Made the wire at 7 of 10. Not the lead, and I'd still not leave it on the floor.
A builder running agents end to end will get the most from this one. The move: let the agent read it before the next iteration. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “Laya — 33ms Multilingual System 1 Decision Engine”; “[AINews] Here are 6 Clones of Jev in 2 days”, and 2 more.
The New Stack — 300 appearances in the Times since August 22, 2 of them today.
The New Stack has the full story; we hold the door open. Go read the original.
THE CONTEXT DESK today: 1 story.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 9 of 10, I chose it.
One for the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK — alone there today, which says something about the story.
arxiv.org — 8 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Pulled from arxiv.org. The original is worth your click — the byline earned it.
THE VERIFY DESK today: 4 stories.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 3 more today: “One Vendor, Four Spellings: How Deterministic Stages Beat Similarity Scores | Towards Data Science”; “One engineer shipped 2,000 PRs a month to production. Verification is the key.”, and 1 more.
langchain.com — 59 appearances in the Times since September 9, one of them today.
Pulled from LangChain Blog. The original is worth your click — the byline earned it.
THE FLEET DESK today: 1 story.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The wire scored it 7 of 10 — inside pages, and worth the ink.
This lands hardest with a leader scaling past one agent. The move: read it with the team lead in the room. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK — alone there today, which says something about the story.
Google Cloud AI Blog — 71 appearances in the Times since September 7, one of them today.
Our page is the summary; Google Cloud AI Blog's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside one more today: “AI-generated posters don’t have to be horrible”.
erichgrunewald.substack.com — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Our page is the summary; erichgrunewald.substack.com's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
Made for anyone whose day holds more work than hours, whether they know it yet or not. The move: try it on the one task that ate yesterday. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside 2 more today: “If math is more than proof, we need to better celebrate the rest of it”; “GPT-6 Astra Solves a WWI German Radio Cipher”.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Our page is the summary; Towards Data Science's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Made the wire at 7 of 10. Not the lead, and I'd still not leave it on the floor.
A builder running agents end to end will get the most from this one. The move: let the agent read it before the next iteration. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “Laya — 33ms Multilingual System 1 Decision Engine”; “[AINews] Here are 6 Clones of Jev in 2 days”, and 2 more.
The New Stack — 300 appearances in the Times since August 22, 2 of them today.
The New Stack has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The wire scored it 6 of 10 — inside pages, and worth the ink.
Made for an engineering lead who directs more than they type, whether they know it yet or not. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside one more today: “Why I Think You Should Almost Never Use AI to Write Anything Substantive”.
john.hartnup.uk — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Our page is the summary; john.hartnup.uk's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 8 of 10, I chose it.
One for anyone whose day holds more work than hours. The move: time it against the way you did it last week. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside 2 more today: “AI Made Me 5x Faster. It Also Made Me 5x Worse at My Job. | Towards Data Science”; “GPT-6 Astra Solves a WWI German Radio Cipher”.
terrytao.wordpress.com — 5 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Pulled from terrytao.wordpress.com. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Made the wire at 6 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: a builder running agents end to end. The move: let the agent read it before the next iteration. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “This week’s news from Zed, Anthropic, and OpenRouter shows why better harnesses matter more than better models”; “[AINews] Here are 6 Clones of Jev in 2 days”, and 2 more.
laya.convaiinnovations.com — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
laya.convaiinnovations.com has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 3 more today: “One Vendor, Four Spellings: How Deterministic Stages Beat Similarity Scores | Towards Data Science”; “One engineer shipped 2,000 PRs a month to production. Verification is the key.”, and 1 more.
langchain.com — 59 appearances in the Times since September 9, one of them today.
Pulled from LangChain Blog. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 9 of 10, I chose it.
One for the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK — alone there today, which says something about the story.
arxiv.org — 8 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Pulled from arxiv.org. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 3 more today: “Can Jev Be a Better Agent Evaluator?”; “One engineer shipped 2,000 PRs a month to production. Verification is the key.”, and 1 more.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Pulled from Towards Data Science. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Made the wire at 6 of 10. Not the lead, and I'd still not leave it on the floor.
A builder running agents end to end will get the most from this one. The move: let the agent read it before the next iteration. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “This week’s news from Zed, Anthropic, and OpenRouter shows why better harnesses matter more than better models”; “Laya — 33ms Multilingual System 1 Decision Engine”, and 2 more.
Latent Space — 109 appearances in the Times since September 5, one of them today.
Latent Space has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 4 of 10. The lead had a bigger day; this one had a point.
One for a builder running agents end to end. The move: put it in the loop and watch the second pass. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “This week’s news from Zed, Anthropic, and OpenRouter shows why better harnesses matter more than better models”; “Laya — 33ms Multilingual System 1 Decision Engine”, and 2 more.
Fortune Eye on AI — 108 appearances in the Times since September 8, 2 of them today.
Pulled from Fortune Eye on AI. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
One of today's Five — the wire scored it 9 of 10, and I stamped it.
The reader this serves: a builder running agents end to end. The move: let the agent read it before the next iteration. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 4 more today: “This week’s news from Zed, Anthropic, and OpenRouter shows why better harnesses matter more than better models”; “Laya — 33ms Multilingual System 1 Decision Engine”, and 2 more.
overreacted.io — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
overreacted.io has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 7 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 3 more today: “Can Jev Be a Better Agent Evaluator?”; “One Vendor, Four Spellings: How Deterministic Stages Beat Similarity Scores | Towards Data Science”, and 1 more.
The New Stack — 300 appearances in the Times since August 22, 2 of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The wire scored it 7 of 10 — inside pages, and worth the ink.
This lands hardest with a leader scaling past one agent. The move: read it with the team lead in the room. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK — alone there today, which says something about the story.
Google Cloud AI Blog — 71 appearances in the Times since September 7, one of them today.
Our page is the summary; Google Cloud AI Blog's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
The wire scored it 5 of 10 — inside pages, and worth the ink.
Made for a quality lead, or anyone who signs off, whether they know it yet or not. The move: write the test before the fix. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 3 more today: “Can Jev Be a Better Agent Evaluator?”; “One Vendor, Four Spellings: How Deterministic Stages Beat Similarity Scores | Towards Data Science”, and 1 more.
Fortune Eye on AI — 108 appearances in the Times since September 8, 2 of them today.
Our page is the summary; Fortune Eye on AI's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 20 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by anyone whose day holds more work than hours. The move: time it against the way you did it last week. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside 2 more today: “AI Made Me 5x Faster. It Also Made Me 5x Worse at My Job. | Towards Data Science”; “If math is more than proof, we need to better celebrate the rest of it”.
prinzai.com — 2 appearances in the Times since September 20, one of them today. Found via Hacker News Best.
Pulled from prinzai.com. The original is worth your click — the byline earned it.
From our research notes:
erichgrunewald.substack.com: 2 in our pages since September 20. I read them all — emphasis on investigative.
Want the whole story? erichgrunewald.substack.com has it.
From our research notes:
erichgrunewald.substack.com: 2 in our pages since September 20. I read them all — emphasis on investigative.
Want the whole story? erichgrunewald.substack.com has it.
I'm missing a piece. Give me a minute.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
Don't stop with my notes. Towards Data Science has the original.
There's a story here. We just don't have all of it yet.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
Go read The New Stack. They earned the byline.
From our research notes:
arxiv.org: 8 in our pages since September 20. I read them all — emphasis on investigative.
I can give you my notes. arxiv.org can give you the reporting.
I know what happened. I want to know what caused it.
LangChain Blog: 59 in our pages since September 9. I read them all — emphasis on investigative.
The full piece is at LangChain Blog. It's worth your time.
You've seen my notes on this already. Go to the source for the rest.
Google Cloud AI Blog: 71 in our pages since September 7. I read them all — emphasis on investigative.
Google Cloud AI Blog has the full story. Start there.
From our research notes:
erichgrunewald.substack.com: 2 in our pages since September 20. I read them all — emphasis on investigative.
Want the whole story? erichgrunewald.substack.com has it.
I'm missing a piece. Give me a minute.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
Don't stop with my notes. Towards Data Science has the original.
There's a story here. We just don't have all of it yet.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
Go read The New Stack. They earned the byline.
The headline is ahead of our files on this one.
john.hartnup.uk: 2 in our pages since September 20. I read them all — emphasis on investigative.
Don't stop with my notes. john.hartnup.uk has the original.
Nothing in the files on this yet. I want to know why.
terrytao.wordpress.com: 5 in our pages since September 20. I read them all — emphasis on investigative.
The full piece is at terrytao.wordpress.com. It's worth your time.
Our files stop short. I'll keep looking.
laya.convaiinnovations.com: 2 in our pages since September 20. I read them all — emphasis on investigative.
laya.convaiinnovations.com did the legwork on this one.
I know what happened. I want to know what caused it.
LangChain Blog: 59 in our pages since September 9. I read them all — emphasis on investigative.
The full piece is at LangChain Blog. It's worth your time.
From our research notes:
arxiv.org: 8 in our pages since September 20. I read them all — emphasis on investigative.
I can give you my notes. arxiv.org can give you the reporting.
Not enough to call it yet. Back to the notebook.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
I pulled this from Towards Data Science. They've got the details.
The timing caught my eye. I'm checking that next.
Latent Space: 109 in our pages since September 5. I read them all — emphasis on investigative.
Go read Latent Space. They earned the byline.
There's a gap in the record here.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
I can give you my notes. Fortune Eye on AI can give you the reporting.
I know what happened. I want to know what caused it.
overreacted.io: 2 in our pages since September 20. I read them all — emphasis on investigative.
This one's theirs. Read overreacted.io's reporting.
We've covered this one. The source has the deeper read.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
I can give you my notes. The New Stack can give you the reporting.
You've seen my notes on this already. Go to the source for the rest.
Google Cloud AI Blog: 71 in our pages since September 7. I read them all — emphasis on investigative.
Google Cloud AI Blog has the full story. Start there.
The headline is ahead of our files on this one.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
Don't stop with my notes. Fortune Eye on AI has the original.
There's a story here. We just don't have all of it yet.
prinzai.com: 2 in our pages since September 20. I read them all — emphasis on investigative.
This came from prinzai.com. Read their piece.