This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
He is a startup founder with a record high burnrate with a product that isn't good enough to be a commercial product. He has to bring in lots of customers promising it will soon be better. He wants to speedrun to a point in which he can actually generate revenue that can cover his costs.
Chatgpt is not good enough to replace developers, lawyers or any other qualified profession. It is an alternative to google without the ability to insert advertising. Github co-pilot isn't great for overall productivity.
Chatgpt is facing scaling laws. The bigger the model the more power it draws. The size of the model required to be useful is too large for today's hardware and power costs. They can no longer make a model orders of magnitude larger. The datasets are too large to fine tune manually.
Sam Altman is pretty much begging for a nuclear reactor and enough GPUs to swallow all that juice.
Also, OpenAI had to raise prices to keep themselves from losing money on ChatGPT.
More options
Context Copy link
I recently asked it to define "Chinook Trough" and it told me directly that nothing of the sort exists. When I explained that it does in fact exist (I spent too much time staring at the skymap over the Pacific) it changed tunes, apologized, and gave me a relatively clear definition. I only mention this as usually ChatGPT has been good at this sort of straight explanation query.
That's weird. Usually LLMs have exactly the opposite problem. I would find them infinitely more useful if the worst case output was "I've never heard of that" rather than confident-but-wrong hallucinations. I guess "that doesn't exist" is still in confident-but-wrong territory, but it's not the usual "yes, and" improv ad-libbing I see.
More options
Context Copy link
More options
Context Copy link
I second this take, full disclosure I'm a middling noob learning programming and math post college who isn't very smart.
Sam has been making similar announcements since they serve as good pr and boost your firms morale, today anthropics models seem to perform better and llms aren't the only thing in AI. One of the people who originally did play role in llms getting where they are is Jery Howard of fast.ai fame who worked on Ulmfit and his comments have been fairly skeptical, I usually lean on his takes for most things ai. I did do the first few lessons on fast.ai before chsggpt blew up, he has a good description of the training process within transformers.
More options
Context Copy link
We'll see I guess. DeepSeek trained a GPT-4 level AI for $6 million (admittedly employing existing LLMs). They have also made huge efficiency gains in inference, charging just $0.14 per million tokens as compared to $3 per million output tokens with a comparable Claude model.
Software is becoming more efficient much more quickly than hardware. We might not need those terawatt scale data centers until after AGI is achieved.
On a theoretical level, absent some sort of woo about quantum computation in the human brain, there's no reason why silicon shouldn’t be vastly superior to synapses eventually.
I'm a huge DeepSeek fan so will clarify.
Those are their own LLMs, and they collectively bump that up to no more than $15M, most likely (we do not yet know the costs of R1 or anything about it, will take a few more weeks; V2.5 is ≈2.2M hours).
0.14/1M input, 0.24/1M output vs $3/$15, to be clear. There are nuances like 0.014 for 1M input in the case of cache hits, opt-in paid caching on Anthropic, and the price hike to come in February.
But crucially, they've published model and paper. This is most likely done because they assume top players already know all these techniques, or are close but work on another set that'll yield the same effect.
More options
Context Copy link
I think the phrase "quantum woo" vastly understates the potential impact of quantum computing on machine learning. The quantum algorithm zoo, for example, lists a number of quantum machine learning algorithms. Several of these get exponential speed up from classical algorithms, but even a quadratic speedup of grover's algorithm would be game changing at the scale frontier models operate on.
I agree that most normie use of quantum in the brain is "woo". And I also agree that it's not been established that the brain relies on any quantum effects. But there is actual legitimate research in these directions and it seems wrong to offhandedly dismiss it.
Viable quantum computing dropping today (or even in the next decade or two) would also break almost all extant (asymmetric) cryptography. Yeah, NIST just recently published specs for post-quantum crypto, but I expect it'll be a decade before those are universally supported. Maybe less if it happens: SSL everywhere happened fairly fast, but became a real priority almost overnight. But if quantum were something any well-founded startup could do, nation-state actors could throw some impressive wrenches into any secure networks for a while.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link