While it's noisy and complicated for humans to read through, this session info is primarily for future AI to read and use as additional input for their tasks.
We could have LLMs ingest all these historical sessions, and use them as context for the current session. Basically treat the current session as an extension of a much, much longer previous session.
Plus, future models might be able to "understand" the limitations of current models, and use the historical session info to identity where the generated code could have deviated from user intention. That might be useful for generating code, or just more efficient analysis by focusing on possible "hotspots", etc.
Basically, it's high time we start capturing any and all human input for future models, especially open source model development, because I'm sure the companies already have a bunch of this kind of data.
That's exactly one of the reasons I've been archiving the sessions using DataClaw. The sessions can contain more useful information than the comments for humans.
TBH I don't think it's worth the context space to do this. I'm skeptical that this would have any meaningful benefits vs just investing in targeted docs, skills, etc.
I already keep a "benchmarks.md" file to track commits and benchmark results + what did/ did not work. I think that's far more concise and helpful than the massive context that was used to get there. And it's useful for a human to read, which I think is good. I prefer things remain maximally beneficial to both humans and AI - disconnects seem to be problematic.
Might not be worth it now, but might be in future. Not just for future LLMs, but future AI architectures.
I don't think the current transformers architecture is the final stop in the architectural breakthroughs we need for "AGI" that mimics human thought process. We've gone through RNN, LSTM, Mamba, Transformers, with an exponentially increasing amounts of data over the years. If we want to use similar "copy human sequences" approaches all the way to AGI, we need to continuously record human thoughts, so to speak (and yes, that makes me really queasy).
So, persisting the session, that's already available in a convenient form for AI, is also about capturing the human reasoning process during the session, and the sometimes inherent heuristics therein. I agree that it's not really useful for humans to read.
I just don't really see the point in hedging like that tbh. I think you could justify almost anything on "it could be useful", but why pay the cost now? Eh.
> While it's noisy and complicated for humans to read through, this session info is primarily for future AI to read and use as additional input for their tasks.
Context rot is very much a thing. May still be for future agents. Dumping tens/hundreds of thousand of trash tokens into context very much worsen the performance of the agent
It's just noise for AI too. There is no reason to be lazy with context management when you can simply ask the AI to write the summary of the session. But even that is hardly useful when AI can just read the source of truth which is the code and committed docs
Q: Will you turn off the tool if they violate the rules?
@sama: Yes, we will turn it off in that very unlikely event, but *we believe the U.S. government is an institution that does its best to follow law and policy.*
What we won't do is turn it off because we disagree with a particular (legal military) decision. We trust their authority.
How can anyone trust a guy who says this after the past year?!
I bet Sam secretly pledged to DoD that the red lines were only temporary, for optics and to calm employees at the all hands meeting.
A few months down the line, OpenAI will quietly decide that their next model is safe enough for autonomous weapons, and remove their safeguard layer. The mass surveillance enablement might be an indirect deal through Palantir.
The fact they were able to strike a deal in the first place hours after Anthropic was declared a supply chain risk should make this obvious. Their employees are smart people, the only way they can reason past this is their compensation.
*Altman may have killed a king by getting him to be complacent.*
I still think a lot about the failed OpenAI coup, and how different things would be now if Microsoft hadn't backed Altman. Would this hype cycle and bubble grown so ridiculous if there were more conscientious people in charge at the front-runner? We will unfortunately never know. I really wish that board had planned out their coup better.
Bingo! Elon's main life mission now is to roll back social progress via the anti-woke combination of xAI and Twitter. That's why he's tying them to the now rather-essential SpaceX, despite possibly hurting its IPO. He can now keep pumping money into them without a worry.
There was a WaPo article yesterday, that talked about how xAI deliberately loosened Grok’s safety guardrails and relaxed restrictions on sexual content in an effort to make the chatbot more engaging and “sticky” for users. xAI employees had to sign new waivers in the summer, and start working with harmful content, in order to train and enable those features.
I assume the raid is hoping to find communications to establish that timeline, maybe internal concerns that were ignored? Also internal metrics that might show they were aware of the problem. External analysts said Grok was generating a CSAM image every minute!!
I think Musk is just that obsessed with his mission of reversing social progress and controlling the direction of the world, using the anti-woke combination of xAI and Twitter. He knows that tying them to SpaceX will hurt its IPO, but now they're part of an entity that's too essential to fail.
They're also probably rushing out the IPO to beat the bubble pop. I think everyone earlier expected to keep the bubble going a few more years, that's why they made all those circular deals. But then Trump spooked Europe into possibly scaling back US investments and decoupling from US tech. So now you have an unsure Nvidia walking back their OpenAI deal, etc.
The "administrative costs" in that analysis are misleading. "Administrative costs include spending on running governmental health programs and overhead from insurers but exclude administrative expenditures from healthcare providers." I don't think the blue bar is representative of the "raw cost" of care.
This study says over one-third of all US healthcare costs are administrative.
We could have LLMs ingest all these historical sessions, and use them as context for the current session. Basically treat the current session as an extension of a much, much longer previous session.
Plus, future models might be able to "understand" the limitations of current models, and use the historical session info to identity where the generated code could have deviated from user intention. That might be useful for generating code, or just more efficient analysis by focusing on possible "hotspots", etc.
Basically, it's high time we start capturing any and all human input for future models, especially open source model development, because I'm sure the companies already have a bunch of this kind of data.
reply