Yeah, this is a perfect microcosm of how "the system" works. This guy is right! He just got there too late, with too little money and not enough fanfare.
Sam Altman, on the other hand? He's the figurehead of a consumer-facing product with a brilliant marketing team and probably a crack team of lobbyists to boot. He's got billions of dollars in Microsoft funding and a product you can go use for free. All he's got to do is roll in there first with an AI alarmist response and convince lawmakers the state-of-the-art is dangerous. After all, OpenAI is synonymous with SOTA, right? So we follow the industry leaders, without really looking where they're headed. And now we end up here.
Oh how I wish Open Source software was fairly represented in court. But ideas don't pay for themselves, and money talks awful loud...
> All he's got to do is roll in there first with an AI alarmist response and convince lawmakers the state-of-the-art is dangerous.
Mr Altman didn't go before congress to convince them of anything, he was invited by Blumenthal to make their case to the media. Congress is embarrassed that social media got out of hand and is chomping at the bit to regulate AI. There was no disagreement at the testimony, the intentions of government and industry are aligned perfectly here - roll out the red carpet for centralized control, regulate competition out of existence.
This sounds correct. My only hope is since AI will be regulated in the US, there is still a chance that given other countries do not have the same leverage as the US, they will keep AI unregulated as a way to avoid being controlled by the US.
AI will not be regulated in the US because the first amendment prohibits such regulation. Software is speech and is this protected per Bernstein v DOJ.
You can distribute binaries as speech but the services you're offering as an API is a different matter, more interstate commerce, what you're allowed to sell as opposed to what you're allowed to say
The government's interests are going to align more with OpenAI's, and they would be fools not to eliminate open source/local models.
Anything local bypasses traversal of any network edges they can tap and monitor. Anything you ask Google is logged, and your search history can be subpoenaed, same as with any public library and the books you check out. OpenAI or whatever gatekeeper would assume that role and charge tolls.
Do the same research with Vicuna-13B-ggml or whatever and nobody can sniff out what you're doing. Which sounds great for people who just want to jerk off and be left alone, but has terrifying implications for anybody tasked with stopping more-malicious users.
Unfortunately nobody can ever compromise on these things. Best case would be adopting a law like the recent one in NL, where given probable cause the government can hack into your machine, but that won't happen. The only alternative in their favor is banning local models and open-source development of them altogether.
People yearn for "a room of their own"[1], a sanctuary for contemplation and creativity. There was a time when our minds and pieces of paper were the only canvases for our imaginations, but increasingly our web browsers have morphed into thought surveillance systems. Now, the tide is turning back towards private creative spaces, enhanced by generative AI. Even if there are attempts to ban it, the edge now lies with private AI. In the future, we might even have the ability to run even more powerful models on standard hardware.
Moreover, we must consider the defensive strategies against other AI entities that will soon permeate the internet. Operating much like a firewall, private AI models could safeguard users from external threats such as involuntary data breaches and misinformation. This will be an endless cat-and-mouse game akin to battling security exploits or viruses—there's no one-size-fits-all solution. Venturing into the internet without your AI shield may soon be as unthinkable as stepping outside without a mask amidst a pandemic like COVID.
I’ve unfortunately not found an open source model that can perform at the this level for anything beyond madlibs-style text generation. Most of my prompts are a combination of sequential tasks and one-shot prompting, typically between 1K-2K tokens and everything I’ve tried falls down at this. Granted, I haven’t gone beyond 30B models, but the gulf still seems quite wide (and even wider compared to GPT-4). Really hope it does close in the next year!
I admit I have not pushed them that far, though I am also using 40B and 65B models which may perform better than a 30B model.
Nothing open approaches GPT-4 yet, which I have heard is an overlay of multiple 220B models.
The only way you will see that on even high end consumer hardware is advances in compression and pruning, but those are happening. There’s a lot of research in those areas.
Of course then cloud models could use the same techniques to get even bigger, if the model and training will keep scaling.
Edit: as AI matures you may also see lots of custom ASICs hit the market. I could see those getting cheap enough to run many hundred B models locally. The acceleration of an ASIC designed for a specific type of model over CPU and GPU could be dramatic, reducing the amount of silicon you need.
In my experience, Wizard-Vicuna output at 13b will match or exceed the prose quality of GPT3.5 in smaller generation tasks. I've been using it for a Discord bot to great success.
There is an argument about the open source models causing more innovation and thus more industry and/or competetive advantage of a nation. So I can't imagine that control is the only thing they are thinking about, though surely they are thinking about control.
The horrible thing is that in the public opinion OpenAI is also incorrectly connected with "Open Source" purely due to the similarities in the name. So the two opposing approaches are conflated in the public view.
In my opinion, it's definitely too late. ChatGPT is one of the most well-marketed products of all time, and it frankly didn't achieve all that much. It scaled up current SOTA and put it behind a paywalled API, and the rest is history. It's like the fountain of youth, a dream API for most developers. Put something in, get something out. Whoever owns the endpoint just prints money.
Everything after this point will just be habitually disappointing. Everything will be held to the impossibly high standard of the first time you tried ChatGPT, not the more realistic benchmarks of GPT-2, LLaMA, Bert, or even Talk-To-Transformer. I don't think it will be possible to recapture that magic, and Sam Altman was the first-mover on turning that awe into fear.
Weird take; doesn't strike me as at all inevitable. I can't think of any similar huge thing that won like this merely by being a first-mover. Google, Facebook, Spotify etc etc?
Eh, I think they flubbed it. Agreed that they put it on the market at a time that made a splash and gave a lot of magical first impressions, but at this point the possibility of ever trusting these things is burned, if improved models manage to lie a little less, that won't be enough to actually integrate them into mission critical tasks.
Instead of an oracle they released an improv comedian. Entertaining enough but not what you want running the show.
Someone will build a better app layer that supersedes theirs and they will have to hope that app sells out. The pioneering innovation left the building when they quit being open imo and their frontend / pace of development on the frontend leaves a lot of opportunity.
I’m optimistic mostly because, a lobbyist for closing ai has the disadvantage that congress isn’t good at doing these days, and the status quo is “do nothing” on this issue. Overcoming the resistance of the status quo requires a large popular support that isn’t there and is very expensive to buy with money.
I don’t want to be that guy but iirc Sam was a wealthy person before OpenAi so it’s likely he had connections which other people didn’t have so there’s that. I could be wrong though.
Open source is an essential building block for a sane digital society that would resemble our professed values.
AI is an important but ultimately small part of this overall transformation. Let me remind everybody that bitcoin, aiming to disrupt the very fundamentals of economic life, is also open source.
The evolutionary process of embedding open source capabilities at societal scale is far from automatic or manifestly benign (see above crypto reference).
A lot of technology, institutions, behaviors still need to be imagined, developed, tested and evangelized. That is the good, roll back your sleeves challenge for current and future generations of digital "social" engineers.
I struggle to see any alternative universe that is not dystopic. Genuinely interested if people have articulated "proprietary" visions that are intellectually honest as to their incentives and morally aligned with ideas of individual agency and freedom and democratic controls.
Their CEO did a fine job! I shared a five minute YouTube link with some friends and family. He did a great job explaining the advantages of open models and open source.
In the case of LLMs, I really wish we would stop referring to models where the source code and/or weights were leaked as "open source". Likewise, for those where they were deliberately released but the license restricts them to "non-commercial use" or allows commercial use so long as the use "does not compete" with the company that released them.
Without getting too hung up on the semantics, what matters here is a free exchange of ideas, algorithms, and indeed some code. All this has massively accelerated progress in this space in the last two decades. Instead of sitting on their results, jealously guarding their secret sauce, etc. companies are sharing a lot of detail and are verifying each other's results by replicating them and improving on them. So OpenAI did chat gpt, Meta did llama, Google is doing Bard. Llama sort of got unleashed on the open source community and people are now starting to use that, improve it, benchmark against it with their own models, etc.
In this case semantics do matter so to keep the improvements in the Software Commons as Commons. Weak / Permissive licenses will ensure that all the voluntarily contributed labour would be enclosed and profitable only to those who can hire an army of engineers. Strong licenses will ensure your call for a free exchange.
Depends. Overly open licenses can also motivate prevent people to not share things. Case in point here is some of the openai stuff where they shared enough so people can replicate their results but without giving away everything. People have been complaining that that's not open enough. But it's still better than them not sharing anything at all. That model certainly seems to be working well for them.
Compare that with e.g. companies in China or elsewhere that are also working on AI that are perhaps sharing a bit less. That's a bit of a black hole in terms of information going in but not coming out. I think the more open people can get the better. But even some weaker licenses are better than nothing.
Things are open enough currently for people to independently replicate each other's results. I think that's the important bit.
Open source is always important. You won't have a good developer pipeline without it. Other engineering disciplines should strive for something similar. Doesn't mean there cannot be closed source software of course.
At least for image synthesis, I believe the open models just plainly produce far better results by now since training is crowdsourced.
I honestly feel the opposite. I mean, the more OSS the better, but I don't think that LLMs being open source will do much to mitigate what I see as being the largest risks of LLMs.
Weird to see huggingface in the news like this. I remember contributing some fix to it back in summer 2019 when it had 4k stars or something, and I thought it was a really popular project even then :)
HF's CEO here! yes hopefully if we can help power good AI democratization thanks to open science and open source, we'd be happy. Do you think that's useful?
Is the huggingface website/hub API itself open-source? That would be a good place to start for people who are interested but don't know a lot about ML. Plus it's some nice goodwill gesture that helps the "brand" in the circles it's talked about.
it's not open-source as it's how we make money to fund all the rest. The huggingchat interface is fully open-source though: https://github.com/huggingface/chat-ui
Thank you for the answer, how successful at converting casual users into paying members?
One of the biggest problems Tech companies face is sustainable scaling, I hope you can scale at a manner that allows you to stay relevant without teetering at the cliff's edge.
I think that is extremely obvious, given their interface looks 90% like github.
They are trying to branch out into training AI models and running inference themselves. But their costs for inferencing look extremely uncompetitive compared to GPU providers like coreweave.
Whatever it is, I really, really, really, really hope it's doesn't fall into the hands of microsoft, google or the rest. (But of course it will and enshittification ensues.)
This provision initially had some meaning, but court cases in the 1940s expanded it to include just about anything that had so much as a hypothetical butterfly effect on interstate commerce - thereby handing Congress nearly unlimited power.
Wait what would a king know about congress? Actually what would a philosopher know about wielding power? Anyway, here's what journalists and think-tanks say about it:
if Congress knew how to work at all, computer laws from the 80s would be updated, they'd work on some sort of hit/git way of submitting bills such that we don't need huge omnibus bills and can singularly insert little things as needed or update and see a full change log.
They'd also ban all politicking outside of an election season on social media, the news, etc and only allow talk about issues affecting the country and Congress that are under their pervue.
They'd also have better rules about etiquette in Congress and if you defy the rule there's a 5k fee, that goes up 1k every time you repeat. Congress should function a bit like a court room.
Filibusters should've never been knee capped, and they should be focused on laws that really benefit Americans, before corporations, not take money from rich donors, etc. For example if Congress worked, A life saving drugs like insulin would be price regulated. At least after the drug is break even with R and D costs, then allow a max 150 percent markup. Current markup is 1700 percent iirc.
If Congress worked they'd figure out a way that everyone can receive guaranteed healthcare. Every other country does it. They don't want to, because insurance lobbies don't want them too and they pay them their weight in gold.
If Congress worked they'd realize the more money people locally have to spend the more money gets spent. fuck, even Henry Ford understood this concept, essentially it's the reason for the 40 hour work week. Without leisure time he postulated that people wouldn't buy cars because they'd be too worn out. He was right, far sales rocketed.
Middle class and lower class generally spend 90 percent of their income and most locally to the country or their own city. Money doesn't trickle down, it flows up.
Congress is contributing to the polarization of America, you can look at congressional sessions over time and they've evolved to this cesspool that Congress is today, just compare 96 with today or 2006 even.
If Congress worked they'd see the value of ranked choice voting. Even if it cost them their job. Even if it led to many third parties winning elections, hell especially this. The founding fathers knew the two party system would be our ruin but didn't have a plan to tackle this, but they cared about our future. Congress cares about today and maybe 2 years from now max, otherwise global warming would have more focus, though I guess tomorrow is catching up with us on that one.
Have you see Bittensor's project and what are your opinions? Would you consider integrating?
I believe it may help with the open source AI movement and projects such as this!
Since GPT3 was able to diagnose a problem patient that multiple Physicians could not solve, I have been concerned that we need these local models faster than ever. (And now the patient doesn't need to take Parkinson's medication before! The family doc misdiagnosed and every specialist tried to find their own solution. Thus why my wife didn't know the diagnosis.)
The AMA/Physician Cartel hates competition, they won't even like Physicians from other countries practice without going through the (regulatory capture) ACGME bottleneck.
LLMs have taken/will overtake a physician's abilities for diagnosis. The math/science will prove that LLMs lead to better outcomes. It seems inevitable.
However, when one person is misdiagnosed(and dies or makes for a sad, emotional story), the cartel will be able to point to it, and outspend everyone else lobbying/bribing to make LLM for medical use illegal... for our safety.
Sam Altman, on the other hand? He's the figurehead of a consumer-facing product with a brilliant marketing team and probably a crack team of lobbyists to boot. He's got billions of dollars in Microsoft funding and a product you can go use for free. All he's got to do is roll in there first with an AI alarmist response and convince lawmakers the state-of-the-art is dangerous. After all, OpenAI is synonymous with SOTA, right? So we follow the industry leaders, without really looking where they're headed. And now we end up here.
Oh how I wish Open Source software was fairly represented in court. But ideas don't pay for themselves, and money talks awful loud...