oh yeah that was obvious when you see who they are and what they do. also, one of the large opensource projects was the lesswrong site lololol
i'm surprised it's as well constructed a study as it is even given that
wonder if that's why they tried and failed to change it to "squiggle maximizing"
the comments are wild "yes good post having my brain taken over by a superintelligent autocomplete is a reasonable concern"
it should be, but he's been actively promoting the stuff
YouTube Video
Click to view this content.
‘an expensive slot machine that outputs slop 98% of the time’
podcast and blog post tomorrow, i am ill
yes that's a clothes hanger at the back i forgot to put in the hall, we only reveal our clean laundry here
all the stock sites are. use case: an image that's almost perfect but you wanna tweak it
LEARN PAINT YOU GHOULS
you should be read up on the gospel of @fasterandworse never shutting up about "Hooked"
the psychology of loot boxes?
yep! https://awful.systems/post/4568900
the book is "Hooked" and it's Don't Build The Torment Nexus I'm Now Providing You A Detailed Blueprint Of
AI->cocaine filter: Cocaine isn’t going to replace you. Someone using cocaine is going to replace you.
this sort of bloody stupid metric is widespread, i've heard about it widely
I dabble in conversational AI for work
yeah this may be the wrong sub for you
these are stock images! Which are surprisingly cheap. By Valeriy Kachaev, who puts stuff up as Studiostoks on a pile of stock image sites. His pics are bizarre and keep being the perfect thing.
ahahaha holy shit. I knew METR smelled a bit like AI doomsday cultists and took money from OpenPhil, but those "open source" projects and engineers? One of them was LessWrong.
Here's a LW site dev whining about the study, he was in it and i think he thinks it was unfair to AI
I think if people are citing in another 3 months time, they'll be making a mistake
dude $NEXT_VERSION will be so cool
so anyway, this study has gone mainstream! It was on CNBC! I urge you not to watch that unless you have a yearning need to know what the normies are hearing about this shit. In summary, they are hearing that AI coding isn't all that actually and may not do what the captains of industry want.
around 2:30 the two talking heads ran out of information and just started incorrecting each other on the fabulous AI future, like the worst work lunchroom debate ever but it's about AI becoming superhuman
the key takeaway for the non techie businessmen and investors who take CNBC seriously ever: the bubble starts not going so great
this is not debate club, per the sidebar
LLMs don't even qualify as autotune. Autotune, at least, still has some connection to work (not necessarily talent).

this podcast guy talks like sephiroth but the title is perfect and the content is largely correct
this user has been removed for commenting without reading the article
being from programming dot dev is just the turd on top
Model Evaluation and Threat Research is an AI research charity that looks into the threat of AI agents! That sounds a bit AI doomsday cult, and they take funding from the AI doomsday cult organisat…

and n=16 handily beats the usual promptfondler n=1
youtube comment:
Zeppelins are here to stay! After all, we still have the technology for rigid airship full of hydrogen cells, you can't unring this bell!
(I have pretty good YT commenters actually. Because I moderate all comments, because this is fuckin AI.)
the dream of robot slaves is too powerful to suppress permanently, but it'll be one heck of an AI winter
When some huge and stupid public chatbot disaster hits the news, the AI pumpers will Kramer into the mentions to say stuff like “you have to admit, AI is here to stay.” Well, no, I don’t. Not unles…

It's a tricky one, because a lot of ML these days is transformer-based, because transformers are unreasonably effective, and many things a transformer can do well, an LLM can do okay at. (e.g. translation, transcription, OCR ...)
A lot of people are super convinced by an impressive demo. We have computers you can just talk to now! That's legit amazing, actually! The whole field of NLP is 80% solved! The other 20% is where it's a lying fuckin idiot and that's probably not fixable ...
ooh good, i'll add that generative AI will likely become as radioactive to the public as crypto is.
it's near as joyous as the tesla subs
if you ever wonder how I write Pivot, it's a bit like this. The thing below is not a written text, it's a script for me to simulate spontaneity, so don't worry about the grammar or wording. But how are the ideas? And what have I missed?
(Imagine the text below with links to previous Pivots where I said a lotta this stuff.)
----
When some huge and stupid public AI disaster hits the news, AI pumpers will dive in to say stuff like “you have to admit, AI is here to stay.”
Well, no I don’t. Not unless you say just what you actually mean, when you say that. Like, what is the claim you’re making? Herpes is here to stay too, but you probably wouldn’t brag about it.
We’re talking about the generative AI stuff here. Chatbots. Image slop generators. That sorta thing.
What they’re really saying is give in. AI like it is right now in the bubble is just a permanent force that will reshape society in its image, so we have to give in to it now and do what the AI pumpers say. You know that’s what they really mean.
We get stuff like this egregious example from the Washington State school system. It stars with “AI is here to stay” then there’s a list of AI stuff to force on the kids assuming all of this works forever just like the biggest hype in the bubble. And that’s not true! [OSPI SLIDE]
If you ask why AI’s here to stay, they'll just recite promotional talking points. So ask them some really pointy questions about details.
Remember that a lot of people are super convinced by one really impressive demo. We have computers you can just talk to naturally now and have a conversation! That's legit amazing, actually! The whole field of natural language processing is 80% solved! The other 20% is where it's a lying idiot and probably can’t be fixed? That’s a bit of a problem in practice. Generative AI is all like that, it’s impressive demos with unfixable problems.
Sometimes they’ll claim chatbots are forever because machine learning works for X-ray scans. If they say that, they don’t know enough to make a coherent claim, and you’re wasting your time.
Grifters will try to use gotchas. Photoshop has AI in it, so you should let me post image slop! Office 365 has AI in it, so if you use Word you might as well be using AI! Spell check’s a kind of AI! These are all real examples. These guys are lying weasels and the correct answer is “go away”. At the least.
Are they saying the technology will surely get better because all technology improves? [WAVE HANDS] Will the hallucinating stop? Then they need evidence for that, cos it sure looks like the tech of generative AI is stuck at the top of its S-curve at 80% useful and hasn’t made any major breakthroughs in a couple of years. [o1 GRAPH] It’s an impressive demo, but the guy saying this will have to bring actual evidence it’s gonna make it to reliable product status. And we have no reason to think so.
Are they saying that OpenAI and its friends, all setting money on fire, will be around forever? Ha, no. That is not economically possible. Look through Ed Zitron’s numbers if you need a bludgeon. [ED Z SLIDE] [Ed Zitron]
These AI companies are machines for taking money from venture capitalists and setting it on fire. The chatbots are just the excuse to do that. The companies just are not sustainable businesses. Maybe after the collapse there’ll be a company that buys the name “OpenAI” and dances around wearing it like a skin.
Are they saying there’s a market for generative AI and so it’ll keep going when the bubble pops? Sure maybe there’ll be a market - but I’ve been saying for a while now, the prices will be 5x or 10x what they are now if it has to pay its way as a business.
Are they saying you can always run a local model at home? Sure, and about 0.0% of chatbot users do that. In 2025, the home models are painfully slow even on a high end box. No normal people are going to do this.
I’ve seen claims that the tools will still exist. I mean sure, the transformer architecture is actually useful for stuff. But mere existence isn’t much of a claim either.
Ya know, technologies linger forever. Crypto is still around serving the all important “crime is legal” market, but it’s radioactive for normal people. If you search for “AI is here to stay” on Twitter, you’ll see the guys who still have Bored Ape NFT avatars. Generative AI has a good chance of becoming as radioactive to the general public as crypto is. They’ll have to start calling the stuff that works “machine learning” again.
So. If someone says "AI is here to stay," nail them down on what the heck the precise claim is they're making. Details. Numbers. What do you mean by being here? What would failure mean? Get them to make their claim properly.
I mean, they won’t answer. They never answer. They never have a claim they can back up. They were just saying promotional mouth noises.
Now, I’ll make a prediction for you, give you an example. When, not if, the venture capitalists and their money pipeline go home and the chatbot prices multiply by ten, the market will collapse. There will be some small providers left. it will be technically not dead yet!! but the bubble will be extremely over. The number of people running an LLM at home will still be negligible.
It’s possible there will be something left after the bubble pops. AI boosters like saying it’s JUST LIKE the dot-com bubble!!! But i haven't really been convinced by the argument "Amazon lost money for years, so if OpenAI just sets money on fire then it must be Amazon."
Will inference costs — 80%-90% of compute load — come down? Sure, they’ll come down eventually. Will it be soon enough? Well, Nvidia’s Blackwell hasn’t been a good chip generation so they’re putting out more of their old generation chips while they try to get Blackwell production up. So it won’t be very soon.
So there you go. If you wanna say “but AI is here to stay!” tell us what you mean in detail. Stick your neck out. Give your reasons.
Anysphere is the startup that produces Cursor, your sort-of dependable vibe coding buddy. You tell Cursor what you’d like and it spits out a complete function! This leads to some spectacular vibe c…

You’ve heard of vibe coding, with, I’m sure, not a little horror. Spin the gacha and make a website! And then you have unfortunate experiences. There are people who get quietly called in to f…

The Independent Publishers Alliance, a group of small UK publishers, has filed a UK competition complaint and an EU antitrust complaint over Google’s AI Overviews in search results. [Press Gazette;…

Last year, Wall Street finance firms started making loans, using video cards as collateral — that famously long-term asset class that solidly maintains its value over, ooh, months. Investment bank…

YouTube Video
Click to view this content.
Outdated science fiction is hobbling our dreams.

Venture capitalists Menlo Ventures have released what purports to be a survey: “2025: The State of Consumer AI”. That is, chatbots. [Menlo Ventures] The subtitle is: “AI’s Consumer Tipping Po…

Microsoft is firing 9,000 workers across the company — 4% of all employees. Before this, they laid off over 6,000 in May and another 300 in June. [CNBC] It’s not that Microsoft is doing badly. In Q…

SWE-Bench Verified by OpenAI tests how well a model can solve real bugs in real Python code from GitHub. These bugs are all public information — so the AI models have almost certainly trained on th…

“Mastering Machine Learning: From Basics to Advanced” by Govindakumar Madhavan was published in April by Springer Nature. It’s $169 as an ebook, or $219 as a hardback. [Springer, archive] Retractio…

An internal Microsoft memo has leaked. It was written by Julia Liuson, president of the Developer Division at Microsoft and GitHub. The memo tells managers to evaluate employees based on how much t…

This is our second week of running Google’s Veo 3 video generator through its paces. So when you’ve read this, go and watch the worked examples. And check out last week’s. [YouTube] Our good friend…

don't just read the text - watch the video, it's the point of the exercise. And I worked all weekend on it. Thank you.
Today, let’s look at two Pitchbook reports — the US VC-Backed M&A Outlook for second quarter 2025 and the 2025 US Venture Capital Outlook: Midyear Update. [Pitchbook; Pitchbook, PDF, archive; P…

Meta has won against Kadrey et al — the authors’ case against Meta training their Llama LLM on the authors’ books, including on pirate copies. [Order, PDF; case docket] Both sides brought motions f…

Much beloved US phone company Verizon has updated its My Verizon app to include 24/7 customer service from AI! It’s got a chatbot based on Google Gemini. [Verizon] Verizon tells The Verge: [Verge] …


the guy who got Elon and Grimes to hook up, destroying Twitter and proceeding to blow up the United States
https://davidgerard.co.uk/