Not sure if I believe in the singularity. I think technology brings about singularities from time to time. Each one puts us on a new plateau which seemed unfathomable from the previous plateau but now looks commonplace. I realize the concept of singularity is that we accelerate unheeded to a point of infinite progress, but do you really think this current generation of LLMs (and this round of VC funding) is going to give me a wetware lab at home where I can give myself custom gene therapies to reverse disease and aging? The technology is cool but what I see at this point is most of its capability is already on display and people are rushing to carve out an ownership stake to secure profit. We already see copyright holders putting limits on LLM providers, safety measures, proprietary models solving the Big Problems. Wouldn’t surprise me if this kind of reverse engineering will be restricted to security researchers and black market ablated models in two years, unfortunately.
There will be plenty open source models that would do reverse engineering, no black market or ablation needed.
The smarter the models get, the bigger the loss for the big 2 when an open source model gets released that doesn't talk down to you. The value of open source increases and closed source falls exponentially.
This is certainly what I am hoping for as well. I’m really curious to see where we are at in a year. We have reached a plateau where the labs clearly have capabilities out of reach of any public model. As far as I understand this is purely constrained by compute resources.
The singularity has nothing to do with diffusion of the tech through society, the models can be superhuman in every capacity and only available inside Anthropic
Everyone mistakes sigmoids for exponentials until they pass the inflection point (see COVID for the classic recent example). I write this comment from a suburban house that exists only because we went through a "transportation singularity" which ended with fuel efficient cars and highways. No one talks about a transport time exponential where we will get to work in 1 second in 10 years time!
I’ve hand-labeled so many freaking datasets over the years for supervised learning tasks, if only to build out a PoC. Would 100% hand this off to an LLM and just fine tune the results. I definitely do not want an LLM in the loop at inference time for anything I could build a simple classifier for.
Llm labelling is so incredibly useful, as is transforming labelled data to other valid but different forms (like where you might train on sentences but also randomly replace days of the week with others).
You can also do stuff like setup active learning but have the labeller be an llm as well to make things more efficient.
Definitely worth it for PoC level things if not full ones.
Edit - also reviewing, so it can run training, test, review failure cases and investigate itself. You can leave them to plough through a few iterations where the early part is the classic time you’d have a few obviously bad sets of results.
The comments are revealing. "Men think...", "women need..." We are caught in a semantic-linguistic trap, debating a false dichotomy and pointing out all the places it has respectively harmed us.
I'd argue there's room for men's issues under (4th-wave, intersectional) feminism. The patriarchy mostly benefits a few at the top, but it has also privileged men disproportionately over at least the last few millennia and if we want to have a real conversation about it we have to have a common set of agreements on which to advance that conversation. We're not there yet as a society, we don't have the language or concepts to make this work on any side of the issue. I guess I'm arguing that some people actually do, but it's where we get into areas of gender deconstruction that start raising sharp objections, which is also part of the semantic-linguistic trap.
I’d draw a distinction between management and technical leadership here.
Management, in the face of a stalemate, is going to address roadblocks from a human perspective. They will ask questions and listen. They will get people to talk and sort out their differences. Management will not weigh in with their own heavy handed technical decision unless they (and their team) has failed at the task of teamwork. Because no matter how good their technical chops, they are perpetually too far removed from the actual problem. Management will not really use the product the team is building. They can’t feel the impact of any one technical decision beyond how it impacts the team.
Technical leadership still has skin in the game. Leadership sets technical direction. They evaluate choices and tradeoffs based on technical merits. They are one of the resources management relies on. Leadership is going to have to live with the consequences of their decisions in their day to day work. They must use and build upon the thing they are leading.
Some people fill both roles. Some teams are lacking elements of one or both. Some teams fill these roles perfectly well in ad-hoc fashions, especially when you get teams of people strong in both areas - by far my favorite work experiences, everyone is exceptional and recognizes it in their peers. Then there is no hierarchy or need for it. That dynamic never lasts long, so we implement structures that try to capture it more broadly and often limp along for the sake of continuing to move forward as a group.
It’s: we spent X billion to train our HUGE model and it’s so good we can make money off it. Please fund us for Y billion to build the HUGER model that will maybe destroy the planet, and if it doesn’t we can’t sell it back to users at profit because of legitimate security concerns. And then we’ll need Z billion to train...
Does anyone see the path to ROI here? There is clearly a profitable business model within these big AI labs, but it’s held captive by messianic 90-page IPO risk statements. I know they want to sell restricted access to the more capable models but the market for those kind of professional services doesn’t even exist yet.
Are there enough people using AI for coding and advanced math and physics who need even better models to justify these values? Because for "normie" non-coding tasks (searching, summarizing, writing) which it seems to me are what the vast majority of users would use, we have ChatGPT at work it defaults to the Luna 6 model and it seems fine as-is today.
I disagree, my company is restricted from doing on-prem non-US model deployment. I appreciated not having to look up where each model was from since I don’t pay attention to the releases of all the models I can’t use and don’t know their names very well.
US Gov won't let Chinese models be hosted on any internal infra, not even supposed to use them for local AI. So there is absolutely a market for US-made or non-Chinese models there.
Do you have a source for this? Genuinely curious, I've seen reports online of MIC contractors already using it extensively on-prem. In fact, one of the issues they bring up is how there is no official guidance right now, so no one really knows what to do.
General guidance we’ve received for deployment on protected gov’t networks the past few years has been any open source, whether software libraries or models, won’t pass security review if there are substantial contributions from Chinese/Russian/etc. nationals. There’s even an exceptions list for “well known software” where there are vetted versions for deployment (nginx).
I don't have a source unfortunately, this is mostly internal directives from my work within NIH. I've had many conversations with security officers and IT about my plans to use local models for various projects, and hackles were always raised in regards to Chinese models.
I also did a short stint with an IT operations company, and we were told Chinese models were a no-go as well from our deployments on DoD hardware.
It’s written in the first person and does not have many of the AI tells that I’ve grown used to, but a couple paragraphs in, I am inclined to agree that this is largely AI generated. It’s like the written version of a monotone voice: it carries on methodically for way too long without any real build up or character.
And now I wonder if I’m just harping on some poor human who needs to spice up their writing a bit. Was there any one thing that gave it away to you?
Yes and no. DS9 plays with things at the edges, much like Ian Banks’ novels deal with the behavior that lies at the edge of the Culture and is part of what permits a utopian society to exist at all. It cuts directly to the question of when and if martial force —or other, more subversive mechanisms of power — must be deployed for protection against a force that threatens the existence of an otherwise non-violent society. I think these are criticisms of visions like Roddenberry’s without necessarily undermining them or missing the point entirely. I also think it’s been an ongoing commentary on modern civilization as much as speculative fiction.
Some of the other series and movies miss the point more dramatically for me by just becoming space opera. Star Trek was always political.
Your point on the organisation that the Culture setup "Special Circumstances" to solve the "Paradox of Tolerance" problem and Starfleet having similar issues is an interesting one.
Though I'd say that Starfleet is in itself too "noble" to solve the problems that Special Circumstances solves in the Culture universe (though Section 31 isn't and they do feature heavily in DS9).
Frankly I think that Banks was smart enough to realise that if he want to have compelling narratives in a universe where you have near godlike AI's and everyone (including them) is nice, you need that edge somewhere.
Asher's universe which has a similar premise to the Culture (having come after it) solves it differently, some of the AI's aren't nice or even nice adjacent which also works.
> Banks was smart enough to realise that [...] you need that edge somewhere.
In fact, pretty much all compelling pro-utopia fiction is almost invariably dealing with the fringes of such utopia - both because it avoids getting bogged down in the details of utopia (and risk valid criticism), and because there is no real substitute for conflict as narrative engine. So inevitably you have to go where the rubber meets the road, so to speak.
It's not a store of value by these metrics. If your primary concern is being able to store assets that cannot be seized or frozen by, say, foreign adversaries, it suddenly becomes much more viable. Occasionally up or down 50% is a lot better than down a guaranteed 100%.
Ironically then foreign adversaries target and hack you via wallet backdoors and such. And because it's such an "unseizable" currency, you don't get it back:)
You're redefining store of value to make it fit. You aren't storing value with bitcoin, you're converting value of fiat into bitcoin at a specific point in time and then later able to pull out some other value of fiat.
Its almost a given that the value you pull out won't be what you put in, likely not close, its either much higher or much lower and that changes frequently over time.
reply