Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI's top startups are barely publishing their research

science.org

Post reply on HN
Print view View original

131–140 of 341 posts

Re: AI's top startups are barely publishing their research

#131

Post by jandrewrogers »

Several things can be true simultaneously:

The process and standards of peer-reviewed publication are tedious. If the objective is efficient communication of knowledge, this is not the process that maximizes that outcome. This process is about credentials and imprimatur, not communication.

Academic publication is largely chasing prestige. The value of prestige accrues more to the individual than the company employing the individual. Incentives for companies reflect this.

Many blogified publications are unqualified slop. This sounds like an indictment but a lot of academic peer-reviewed research is also low-quality slop. In many domains, slop is rewarded, devaluing the contributions of people publishing more serious research, which affects incentives. The signal-to-noise ratio has been poor for decades no matter where you get your knowledge.

Any good research you publish will be exploited against you in myriad ways. It is effectively impossible to assert any IP rights. Research turns into a pure cost center if you treat it this way. Companies recognize this and have adapted to reflect it. In some computer science domains the state-of-the-art has been buried in trade secrets for decades such that the academic literature is embarrassingly obsolete.

Publication costs time and money. Given all of the above, why would you invest productive capacity in public communication of research results? The researchers want to work on interesting problems, the companies want to maximize the leverage from their research. Publishing delivers neither.

A lot of fundamental research is done for the purpose of solving a problem, not publishing a paper. Their motivation and personal reward is solving the problem, not publishing it. Once they've solved it, it is no longer interesting and they are on to the next interesting problem.

Some research is coupled with national security considerations. If you do frontier research this is always sitting in the background.

All of this is a consequence of incentives. A large percentage of all basic research no longer happens in academia. I've greatly enjoyed participating in non-publishing research programs. I also understand why they don't publish. I've done a significant amount of interesting foundational research under my own auspices where it was not worth the effort to publish. I'm not chasing prestige, I had a problem I needed to solve.

No one should be surprised by any of this.

Re: AI's top startups are barely publishing their research

#132

Post by geysersam »

dumpsterdiver wrote: post #70
TimCTRL wrote: post #2

Yet none of them would have been here if Google hadn't published "Attention is all you need", the irony.

The irony may only just be unfolding - check out this paper from 2015: https://research.google/pubs/distilling-the-knowledge-in-a-n... # LLM generated summary of the implied irony Google may consider the standalone frontier-model arms race economically irrational, while still considering frontier-model capability strategically indispensable. Its longer game is probably not to avoid building the biggest models, but to…

I think the answer is much simpler: They didn't recognize the potential of that specific architecture. I don't think anyone could have. They valued the perk to employees being able to publish research, and the prestige that followed, and googles ability to attract talent, more than they did the content of those papers.

Re: AI's top startups are barely publishing their research

#133

Post by ReactiveJelly »

teleforce wrote: post #123
sillysaurusx wrote: post #115

Earlier quoted context omitted.

That's also a reason the big labs stopped. Publishing is most valuable to people who have no other way to get the attention of smart strangers. Once you can hire nearly anyone and everyone already returns your calls, the main remaining effect of publishing is to tell your competitors which things worked. This is what happens to every field as it turns from a science into an industry. Chemists published freely until d…

Which is a very ironic and selfish situation when your business model dependent mostly on model training based on available published data, academic and non-academic.

That's why we should enforce copyleft

Re: AI's top startups are barely publishing their research

#134

Post by ninjahawk1 »

colingauvin wrote: post #125
ninjahawk1 wrote: post #90

I can’t speak for other startups, but I applied to the most recent YC batch with my idea for making AI proactive instead of reactive, and pre-being selected I’ve published a paper on recursive self-improvement mapped to the Epoch AI data. I contacted a professor from a university in the UK and he responded since he was working on similar work, then asked me if I wanted to meet with him. We talked for about an hour si…

Curious what you mean by proactive? Could you share a bit more?

Happily, current AI is interacted with in a reactive loop. I open the Claude app, CLI, whatever, say my prompt, get an output.

I personally wanted an AI that was able to reach out to me about my life before I had to reach out to it. An example, a friend just emailed me asking to meet for at 1pm but I have class at 1:30, so a proactive AI would see that conflict and send me a notification about it, asking if the proposed email it drafted works, then I press send.

My personal setup tracks my mouse movement, keyboard, what’s on my screen, and keeps track of what I’m working on through files on my PC. It can update the backend and then restart it on it’s own, meaning I can develop the thing itself while being away from my PC.

The capabilities are more than what I’ve listed, but I want to avoid being too preachy about something I made. Here’s the repo if you wanted to take a look, it’s open-source and connects to the iPhone app:

https://github.com/getorb/Orb-Backend

Re: AI's top startups are barely publishing their research

#135

Post by 11101010010001 »

xtracto wrote: post #77
raincole wrote: post #21

Do startups in other fields constantly publish their research?

That was my first thought. In my experience, particularly with trading firms (the ones doing real science based prediction trading, not TA bullshit) do not publish ANYTHING at all, and haven't for 40+ years. Think Renaissance Technologies or similar.

publications would likely lead to regulation, if the SEC was properly resourced.

Re: AI's top startups are barely publishing their research

#136

Post by jandrewrogers »

noosphr wrote: post #66

I've been at two startups that have done genuine world first fundamental research. The first tried to publish novel results for 3 years in tier 1 journals before finally doing a preprint and telling the tier one publishers to jump in a fire. The second, and ongoing, isn't publishing anything because of my experience with the first. That and avoiding openAI and Anthropic copying our results and leaving us with nothing…

There is almost no benefit in publishing frontier research as a startup. People can try to argue it but it is not defensible. Publishing that kind of thing is a flex that companies risking nothing can do.

Cool for Google. Problematic if you are a startup.

Re: AI's top startups are barely publishing their research

#137

Post by r3trohack3r »

cortesoft wrote: post #89
noosphr wrote: post #66

I've been at two startups that have done genuine world first fundamental research. The first tried to publish novel results for 3 years in tier 1 journals before finally doing a preprint and telling the tier one publishers to jump in a fire. The second, and ongoing, isn't publishing anything because of my experience with the first. That and avoiding openAI and Anthropic copying our results and leaving us with nothing…

> That and avoiding openAI and Anthropic copying our results and leaving us with nothing to show for six months of work. Isn't the POINT of publishing research because you want others to copy it?

If you want people to copy it, but want a 20 year temporary monopoly, you can publish an open source step-by-step guide for someone to recreate what you’re doing with the patent office.

Re: AI's top startups are barely publishing their research

#138

Post by mikepurvis »

ninjahawk1 wrote: post #134
colingauvin wrote: post #125

Earlier quoted context omitted.

Curious what you mean by proactive? Could you share a bit more?

Happily, current AI is interacted with in a reactive loop. I open the Claude app, CLI, whatever, say my prompt, get an output. I personally wanted an AI that was able to reach out to me about my life before I had to reach out to it. An example, a friend just emailed me asking to meet for at 1pm but I have class at 1:30, so a proactive AI would see that conflict and send me a notification about it, asking if the propo…

I'm very interested in this kind of thing as a kind of ADHD brain augment, like it's monitoring my slack, github, email, calendar, active terminals, etc, and helps me prioritize what I should work on as well as weighing whether this or that ping is worth interrupting me for.

I assumed that's what openclaw basically was, but is Orb different from that? And is it fundamentally a different model from the request/response, or is it just request/response in an autonomous loop?

Re: AI's top startups are barely publishing their research

#139

Post by jandrewrogers »

r3trohack3r wrote: post #137
cortesoft wrote: post #89

Earlier quoted context omitted.

> That and avoiding openAI and Anthropic copying our results and leaving us with nothing to show for six months of work. Isn't the POINT of publishing research because you want others to copy it?

If you want people to copy it, but want a 20 year temporary monopoly, you can publish an open source step-by-step guide for someone to recreate what you’re doing with the patent office.

The patents are effectively unenforcible. People stopped patenting frontier computer science research 20 years ago. That is the current reality.

Re: AI's top startups are barely publishing their research

#140

Post by ninjahawk1 »

mikepurvis wrote: post #138
ninjahawk1 wrote: post #134

Earlier quoted context omitted.

Happily, current AI is interacted with in a reactive loop. I open the Claude app, CLI, whatever, say my prompt, get an output. I personally wanted an AI that was able to reach out to me about my life before I had to reach out to it. An example, a friend just emailed me asking to meet for at 1pm but I have class at 1:30, so a proactive AI would see that conflict and send me a notification about it, asking if the propo…

I'm very interested in this kind of thing as a kind of ADHD brain augment, like it's monitoring my slack, github, email, calendar, active terminals, etc, and helps me prioritize what I should work on as well as weighing whether this or that ping is worth interrupting me for. I assumed that's what openclaw basically was, but is Orb different from that? And is it fundamentally a different model from the request/respons…

Exactly what my thoughts were when I first heard about Openclaw, that’s the exact idea of Orb that you pointed out. Letting you make less decisions, right now AI gives you answers but still requires decisions based on the outputs it gives you. This would deepen the actual ability of agents in those channels you listed.

On a fundamental level the backend was designed to do as little LLM calls as possible, for instance it’ll do scans of my screen every 15 seconds, log what’s on it and what’s going on, and store it in a local database, then Orb reviews the entire database every 6 hours for me. Then it’ll schedule wakeups for itself throughout the day, up to 4 so it doesn’t waste my tokens, and schedule notifications based on the last database dump it made.

I have my Claude Code, Codex, and Grok Build all useable by using the “claude -p; codex -p…etc” so you can also use multiple CLI’s in conjunction at the same time on different projects or the same project.

So your question about a loop is kind of right, but it really just collects your data all day and stores it locally on your PC then calls the LLM of your choice and it reviews all the data and makes those proactive moves we’ve discussed. You could theoretically get it to always be scanning by an LLM but that would be a drastic waste of money from what I’ve seen since most things don’t require a call.

Post reply on HN