Anonymous ML Engineer Claims Frontier Labs Hitting Scaling Limits, Facing Existential Threat
An unverified letter from someone claiming to work at a major AI lab warns of internal crises, open-source competition, and that LLMs cannot reach AGI.
What to know
- An unverified anonymous letter claims insiders at a frontier AI lab warn of hitting scaling limits, executive departures, and collapse due to open-source competition—but provides no verifiable specifics.
- The letter argues LLMs cannot mathematically reach AGI and that proprietary models face displacement by open-weights alternatives within a year, mirroring what happened to proprietary image-generation tools.
- Critics note the claims lack concrete evidence and could reflect common public knowledge rather than insider information, while the blogger defends authenticity based on a personal reference in the email.
- Commenters debate whether companies must invest in AI despite risks to avoid being left behind if the technology succeeds, or can safely adopt it later as a mature technology.
The dispute Whether the business logic of herd-following AI investment dominates, or whether waiting for mature adoption offers safer returns. · positions read across 13 posts and comments
The letter lacks verifiable specifics and reads like common knowledge or an LLM-generated response, making it unreliable as insider testimony.
-
“With no actual details, with no information that's not publicly available, and with no actual proof that the author is who he says he is, it might as well be an LLM prompt”
jfloren · Lobsters ↗
Companies rationally must invest in AI hype even if uncertain, because being the only non-investor left behind when technology succeeds carries greater risk than crashing alongside competitors.
-
“You actually want to follow the herd. Like when everybody is crashing but you didn't buy it, so you think you are in better position but, suddenly, the government decides to help the crashed companies and you cannot benefit from it.”
ploum · Lobsters ↗
The letter's authenticity is supported by a personal game reference that would be difficult for an LLM to generate, validating it as genuine insider communication despite vagueness.
-
“there was one final line in the email that I cut off, that convinces me that it's was a specifically targeted and human-written mail.”
nemin · Lobsters ↗
Companies can wait to adopt AI as a mature technology rather than investing during hype, gaining 'second-move advantage' like successful post-bubble web adoption.
-
“It's much cheaper to adopt a mature technology than an early one.”
david_chisnall · Lobsters ↗
Anonymous ML engineer Self-described employee at frontier AI labnemin Blog author and publisherjfloren Lobsters commenterploum Lobsters commenter
How it unfolded 5 developments, newest first · click a bar or a number to jump postscomments
-
5
Blog author defends letter's authenticity based on specific personal reference
Nemin, the blogger who published the letter, responds to skeptics by noting the emailer included a final line referencing a game Nemin is tangentially involved in, which he argues is beyond an LLM's capability to generate.
“there was one final line in the email that I cut off, that convinces me that it's was a specifically targeted and human-written mail.”
— nemin -
Look, I actually buy a lot of what the letter says, but it's also something I could have written myself with my passing knowledge of current AI tech. "They're hitting a wall" isn't exactly a groundbreaking proposition, and the letter offers nothing to back up the assertion. "AI is widely astroturfed" is certainly believable, but again, there's no…
2 more of the top 3 · 8 posts in this stretch
-
> but it's also something I could have written myself with my passing knowledge of current AI tech. Yes, I touch on this in detail in the final section of the post. I think I approached the letter with a healthy amount of skepticism. > it might as well be an LLM prompt: You'll have to take my word for it, but there was one final line in the email…
-
All of this is predicated on the idea that being a late mover is a bad idea. That appears in a lot of management texts but more recently there's a lot more discussion of 'second move advantage'. It's much cheaper to adopt a mature technology than an early one. Take the web, for example. During the .com boom, every business decided it needed to…
-
-
4
Commenters debate whether companies must invest in AI hype despite risks
Discussion emerges around the business logic of AI investment: whether companies must participate in hype to avoid being left behind, even if the technology fails. Views split between risk-management arguments and skepticism of herd-following.
“You actually want to follow the herd. Like when everybody is crashing but you didn't buy it, so you think you are in better position but, suddenly, the government decides to help the crashed companies and you cannot benefit from it.”
— ploum -
You should add that, sometimes, being the only winner is not a solution. You actually want to follow the herd. Like when everybody is crashing but you didn’t buy it, so you think you are in better position but, suddenly, the government decides to help the crashed companies and you cannot benefit from it. And that’s only considering the company…
2 more of the top 3 · 4 posts in this stretch
-
> AI thrives and you didn't buy in: RIP. But... why? If AI becomes much better in 6mo, why not buy in then, rather than having spent 6mo wrestling with now-outmoded technology and practices?
-
I reckon it's all about risk management. **AI crashes and you bought in:** Sucks for everyone, but unless you've ridiculously overcommitted, you're *relatively* in the same neck of the woods as everyone else. **AI crashes and you didn't buy in:** You're in a better position than others, assuming positives coming from the amount of funding and…
-
-
3
Commenters challenge verifiability and specificity of anonymous letter's claims
Lobsters users express skepticism about the letter's authenticity and evidentiary value. jfloren notes the claims lack specifics and could be generated by an LLM prompt, while others question whether the insider knowledge justifies the vague assertions.
“With no actual details, with no information that's not publicly available, and with no actual proof that the author is who he says he is, it might as well be an LLM prompt…”
— jfloren -
>And I'm pretty sure any larger company's upper management is expected to buy into the Current Thing, because if it turns out to deliver, you don't want to be the chump who was left out by not jumping on the bandwagon. This has always felt weird to me, honestly, because it's an extremely safe position to take when your shareholders are usually…
-
-
2
Letter alleges AI discourse poisoned by sponsored influencers and astroturfed hype
The engineer claims voices driving AI hype, including online course influencers, are now sponsored by AI companies. They allege engineers deploy agents as cover projects to market models and scrape data.
“The online discussion is poisoned. A lot of the voices driving the hype, the influencers who sold those online courses, are now just being sponsored by the AI companies themselves.”
— Anonymous ML engineer -
background
Letter argues AGI is mathematically impossible with LLMs — The engineer claims a paper using Cantor's diagonalization argument proves AGI cannot be achieved with LLMs, and references a 2023 leaked Google memo stating 'We have no moat, and neither does OpenAI.'
-
background
Letter claims frontier labs lack competitive moat as open models displace proprietary systems — The anonymous engineer warns that proprietary diffusion models were replaced by open-weights alternatives, image/video generation sales collapsed, and LLMs face the same fate. They argue open-weights models will match or exceed proprietary offerings within a year.
-
1
Anonymous ML engineer letter published claiming frontier labs face collapse
A blog post by Nemin publishes an anonymous email purporting to be from an ML engineer at a frontier AI lab. The engineer claims the company has hit scaling limits on programming tasks, experienced executive departures, and faces IPO stress.
“I work at a frontier AI lab as an MLE. I help train models and do RLHF for frontier stuff at one of the big companies you probably know.”
— Anonymous ML engineer
What people are saying 5 voices from 1 site · best of 13 · verbatim
- Sep 17
-
Given that Symbian was also open, and strictly speaking was growing faster in the installation base than Android + iOS combined (in part because it fit weaker devices), and you mention it was flexible enough for that contract about Office to make any sense at all, I think it's Symbian that was the freshly burning platform. Doesn't matter in the…
-
> That’s also why they are so unethical and selling bad products: at worse, people die, there’s a scandal and the company pays a fine. But individuals have no responsibilty (just remember VW CEO during dieselgate). True professionals run the company into the ground with bad products, sell the company for peanuts to their immediately-previous…
- Sep 16
-
> There's news about MS and Uber cutting back on their token quotas, because they've been spending so much money on them. I think all of Big Tech is shovelling bucketfuls of cash onto this every second, because they somehow still expect that to be cheaper than "being left out". I'm talking more about how all the AI in a lot of products are, at…
- Sep 15
-
And you can even track and experiment now, without any sort of big buy in. I've always been a fan of this with vendors, too... even when I'm happy w/ a vendor I use, I keep loose contact with their competitors and occasionally do a free trial of a product or spend a bit of money on a small install just to keep up and keep options open.
-
I don't think your general (as in, individuals, pension fund etc) investors are all about risk taking (that's more for venture capitalists and startups). They value sustained growth over big all-or-nothing attempts. And I would note that by not investing in the current hype, you are sure to lose existing customers to competitors who *are*…