HNHacker News
TopNewBestAskShowJobs

etaioinshrdlu

6,133 karma · joined November 3, 2016

submissionscomments
etaioinshrdlu··on LLM Inevitabilism
I've operated a top ~20 LLM service for over 2 years, very comfortably profitably with ads. As for the pure costs you can measure the cost of getting an LLM answer from say, OpenAI, and the equivalent search query from Bing/Google/Exa will cost over 10x more...
etaioinshrdlu··on LLM Inevitabilism
This is wrong because LLMs are cheap enough to run profitably on ads alone (search style or banner ad style) for over 2 years now. And they are getting cheaper over time for the same quality.

It is even cheaper to serve an LLM answer than call a web search API!

Zero chance all the users evaporate unless something much better comes along, or the tech is banned, etc...

etaioinshrdlu··on Anthropic signs a $200M deal with the Department of Defense
LLMs are a key enabling technology to extract real insights from the enormous amount of surveillance data the USA captures. I think it's not an understatement to say we are entering a new era here!

Previously, the data may have been collected, but there was so much that effectively, on average no one was "looking" at it. Now it can all be looked at.

etaioinshrdlu··on OpenAI delays launch of open-weight model
It's worth remembering that the safety constraints can be successfully removed, as demonstrated by uncensored fine-tunes of Llama.
etaioinshrdlu··on Ask HN: Who is hiring? (July 2025)
Shit, i've notified the team. We will fix.
etaioinshrdlu··on Ask HN: Who is hiring? (July 2025)
DeepAI | Software Engineer (Full Stack, Django, AWS) | Remote | Contract (W2 or 1099) | Ongoing

DeepAI is the all-in-one creative AI platform: generate images, edit photos, chat with AI (web browsing, voice), create videos and music, all in your browser. We focus on accessibility, affordability, and fun.

What you'll do: - Build and upgrade core features (chat, image, video, etc) - Work full stack: Django, AWS, MySQL, vanilla JS (no frameworks) - Mix of new features and legacy code

You: - 3+ years coding professionally - Strong Python/Django - Comfortable with vanilla JS - AWS/cloud experience - Bonus: Docker, MySQL, LLM APIs, AI/ML interest

Team: - 10 people, work directly with founders - Fun, creative culture - Flexible, ongoing contract (W2 or 1099)

Compensation: - Usually hourly, $50-200/hour. Equity or full time is possible.

To apply: Email team@deepai.org with resume, GitHub, and a short note (mention HN).

Note, we're also hiring for technical customer support roles.

etaioinshrdlu··on Generative AI's crippling failure to induce robust models of the world
I don't think it's accurate anymore to say LLMs are just really good word predictors. Especially in the last year, they are trained with reinforcement learning to solve specific problems. They are functions that predict next tokens, but the function they are trained to approximate doesn't have to be just plain internet text.
etaioinshrdlu··on 10 out of 12 people cured in Type 1 Diabetes Study
And one of the patients in the study died of meningitis...
etaioinshrdlu··on Show HN: Chawan TUI web browser
I rabbit-holed a little and apparently Chrome and Safari no longer even fully pass Acid2 and Acid3?
etaioinshrdlu··on Text-to-LoRA: Hypernetwork that generates task-specific LLM adapters (LoRAs)
What is such a thing good for?
etaioinshrdlu··on Domains I Love
ch.at is nice
etaioinshrdlu··on Launch HN: Tinfoil (YC X25): Verifiable Privacy for Cloud AI
Could you talk more about how how this works? I don't think linked article doesn't given enough detail on how the trust boundary extends from CPU to GPU.

Does the CPU have the ability to see unencrypted data?

etaioinshrdlu··on Launch HN: Tinfoil (YC X25): Verifiable Privacy for Cloud AI
Does the secure enclave also perform the TLS encryption on data leaving the enclave?

Also, if you're decoding TLS on the enclave, wouldn't that imply that you're parsing HTTP and JSON on the GPU itself? Very interesting if true.

etaioinshrdlu··on In 2025, venture capital can't pretend everything is fine any more
Sure, but I think COVID disruptions broadly are a much better explanation for inflation than the previous decade of ZIRP, which the article asserts.
etaioinshrdlu··on In 2025, venture capital can't pretend everything is fine any more
In the chart in the article: https://pivot-to-ai.com/wp-content/uploads/2025/05/zirp-2005...

You can clearly see non-zero interest rates from 2016 to 2020...

etaioinshrdlu··on In 2025, venture capital can't pretend everything is fine any more
Ignoring politics, which I think makes this article somewhat off-topic, I would argue with many of the points in this article - but just one would be that ZIRP specifically caused inflation and required higher interest rates.

The way I remember it, rates were already raised substantially, during the first Trump admin, not due to inflation, lowered again during COVID, and raised due to inflation, where a likely causative factor in the inflation being high government spending and stimulus.

etaioinshrdlu··on AI hallucinations are getting worse – and they're here to stay
The creators are definitely trying to make them tell the truth. They optimize for benchmarks where truthful answering gets a higher score. All the big LLM vendors now have APIs that can ground their answers in search results.

Just because it's a hard unsolved problem, I don't understand the impulse to assert the AI industry is on a war with truth!

etaioinshrdlu··on Era of U.S. dollar may be winding down
I thought the reason we are able to borrow cheaply, and ratings, is basically due to credit-worthiness, and not dollar as a reserve dominance. After all, during the debt ceiling crises, that's when our credit worthiness and ratings were downgraded, but it had nothing to do with the actual currency.
etaioinshrdlu··on Era of U.S. dollar may be winding down
Why does this really matter so much?

I think what matters more is national wealth, wealth per capita, and their growth rates. This is an interesting list: https://en.wikipedia.org/wiki/List_of_countries_by_total_wea...

etaioinshrdlu··on 21 GB/s CSV Parsing Using SIMD on AMD 9950X
Well, Itanium might be a counterexample, they probably tried to make that work for far too long..
etaioinshrdlu··on DoorDash to acquire Deliveroo
If you have more money than time, and you don't mind the quality, it seems like not a bad trade.
etaioinshrdlu··on New Study: Waymo is reducing serious crashes and making streets safer
While Waymo is good for safety, it is constantly collecting data not just from users, but from everyone nearby. The vehicles essentially function as mobile surveillance devices, recording bystanders without consent, with no clear policies on data use, retention, or oversight.
etaioinshrdlu··on DeepMind releases Lyria 2 music generation model
These are actually just the problems easier for researchers to solve, mostly due to a lot of readily available data.
etaioinshrdlu··on The New Moat: Memory
Does anyone really like and enjoy LLM products with memory at this point? To me this seems to be a case where the technical ability to do memory vastly exceeds its actual usefulness (for most people).
etaioinshrdlu··on "A computer can never be held accountable"
I suspect within our lifetimes people will grant AI and robots rights, but with rights come responsibilities, and finally we will be able to hold the computer accountable!
etaioinshrdlu··on Gradual Disempowerment: How Even Incremental AI Progress Poses Existential Risks
Bureaucratic systems have been able to fail like this for a long time: https://news.ycombinator.com/item?id=17350645

Now we have the tools to make it happen more often.

etaioinshrdlu··on DeepSeek releases Janus Pro, a text-to-image generator [pdf]
I'm pretty sure that's still the same multimodal LLM, and considered a form of prompting?
etaioinshrdlu··on DeepSeek releases Janus Pro, a text-to-image generator [pdf]
I would say it's more like a core feature of the latest generation of LLM's, which can be prompted with images+text, and output images+text, along with possibly audio and video.
etaioinshrdlu··on Show HN: Browse Anime from the Terminal
Can someone explain the appeal of cramming what is effectively a GUI into a terminal? This sort of thing has always been popular around here, but I never saw the appeal. The main benefits appear to be usage over SSH, and a certain obscure style due to running in a terminal.
etaioinshrdlu··on DeepSeek releases Janus Pro, a text-to-image generator [pdf]
It doesn't look like it allows editing, which I found surprising for this architecture, as the recent generation of flagship LLMs mostly have this capability (GPT-4o).
← PreviousPage 2 of 34Next →