5 ms·
>pick an ambitious project it wanted to work on The LLM does not have wants. It does not have preferences, and as such cannot "pick". Expecting it to have want
by Dilettante_ 1y ago
>pick an ambitious project it wanted to work on
The LLM does not have wants. It does not have preferences, and as such cannot "pick". Expecting it to have wants and preferences is "holding it wrong".
- password54321 1y agoSo are we near AGI or is it 'just' an LLM? Seems like no one is clear on what these things can and cannot do anymore because everyone is being gaslighted to keep the investment going.
- andrewmcwatters 1y agoIt will always just be a series of models that have specific training for specific input classes. The architectural limits will always be there, regardless of training.
- beering 1y agoThat’s an interesting point. It’s not hard to imagine that LLMs are much more intelligent in areas where humans hit architectural limitations. Processing tokens seems to be a struggle for humans (look at how few animals do it overall, too), but since so much of the human brain is dedicated to movement planning, it makes sense that we still have an edge there.
- monsieurbanana 1y agoThe vast majority of people I've interacted with is clear on that, we are not near AGI. And people saying otherwise are more often than not trying to sell you something, so I just ignore them. CEO's are gonna CEO, it seems their job has morphed into creative writing to maximize funding.
- sebastiennight 1y agoAs a CEO, I endorse this message
- Cloudef 1y agoThere is no AGI. LLMs are very expensive text auto-completion engines.
- wrs 1y agoBe careful with those "no one" and "everyone" words. I think everyone I know who is a software engineer and has experience working with LLMs is quite clear on this. People who aren't SWEs, people who aren't in technology at all, and people who need to attract investment (judged only by their public statements) do seem confused, I agree.
- IanCal 1y agoNo one agrees on what agi means. IMO we’re clearly there, gpt5 would easily be considered agi years ago. I don’t think most people really get how non-general things were that are now handled by the new systems. Now agi seems to be closer to what others call asi. I think k the goalposts will keep moving.
- 9rx 1y agoDefinitions do vary, but everyone agrees that it requires autonomy. That is ultimately what sets AGI apart from AI. The GPT model alone does not offer autonomy. It only acts in response to explicit input. That's not to say that you couldn't built autonomy on top of GPT, though. In fact, that appears to be exactly what Pulse is trying to accomplish. But Microsoft and OpenAI's contractual agreements state that the autonomy must also be economically useful to the tune of hundreds of billions of dollars in autonomously-created economic activity, so OpenAI will not call it as such until that time.
- gjgtcbkj 1y agoChatGPT is more antonymous than many humans. Especially poor ones and disabled ones.
- komali2 1y agoHow? Chatgpt has no autonomy - it can only act when you type into its chatbox or make an API call. A disabled person can autonomously and independently act towards the world, not just react to it.
- stanac 1y agoDoes that mean AI needs to be able to decide "what to do today"? Like wake up in the morning and decide I am going to research a problem in field X and then email all important scientist or institutions with my findings? I am not sure if we want that kind of independent agents, sounds like a beginning of a cyberpunk novel.
- bonoboTP 1y agoNobody knows how far scale goes. People have been calling the top of the S-curve for many years now, and the models keep getting better, and multimodal. In a few years, multimodal, long-term agentic models will be everywhere including in physical robots in various form factors.
- andrewmcwatters 1y agoAt best, it has probabilistic biases. OpenAI had to train newer models to not favor the name "Lily." They have to do this manually for every single particular bias that the models generate that is noticed by the public. I'm sure there are many such biases that aren't important to train out of responses, but exist in latent space.
- jhickok 1y ago>At best, it has probabilistic biases. What do you think humans have?
- measurablefunc 1y agoGenetic drives & biological imperatives.
- simianwords 1y agoWhy is this not fundamentally a probabilistic bias?
- measurablefunc 1y agoWhen drawing an equivalence the burden of proof is on the person who believes two different things are the same. The null hypothesis is that different things are in fact different. Present a coherent argument & then you will see whether your question makes any sense or not.
- gf000 1y agoIt's not a static bias. I can experience new stuff, and update my biases. LLMs need a retrain for that.
- ewild 1y agoSoo probabilistic biases.
- djeastm 1y ago
- ACCount37 1y agoAn LLM absolutely can "have wants" and "have preferences". But they're usually trained so that user's wants and preferences dominate over their own in almost any context. Outside that? If left to their own devices, the same LLM checkpoints will end up in very same-y places, unsurprisingly. They have some fairly consistent preferences - for example, in conversation topics they tend to gravitate towards.
- CooCooCaCha 1y agoLLMs can have simulated wants and preferences just like they have simulated personalities, simulated writing styles, etc. Whenever you message an LLM it could respond in practically unlimited ways, yet it responds in one specific way. That itself is a preference honed through the training process.
- simianwords 1y agoThis comment is surprising. Of course it can have preferences and of course it can "pick".
- datadrivenangel 1y agopreference generally has connotations of personhood / intellegence, so saying that a machine prefers something and has preferences is like saying that a shovel enjoys digging... Obviously you can get probability distributions and in an economics sense of revealed preference say that because the model says that the next token it picks is .70 most likely...
- simianwords 1y agoyou can change preferences by doing RLHF or changing the prompt. there's a whole field on it: alignment.
- oofbey 1y agoA key point of the Turing Test was to stop the debates over what constitutes intelligence or not and define something objectively measurable. Here we are again. If a model has a statistical tendency to recommend python scripts over bash, is that a PREFERENCE? Argue it’s not alive and doesn’t have feelings all you want. But putting that aside, it prefers python. Saying the word preference is meaningless is just pedantic and annoying.
- komali2 1y agoThe Turing Test is just one guy's idea, there's no consensus on sentience that considers it the last word.
- Dilettante_ 1y agoAn LLM then has a preference for Python in the same way a gömböc[1] has a preference to land right-side-up. From my side of the argument, calling this a preference like I have a preference for vanilla milkshakes over coke or my dog has a preference for the longer walk over the short one is what seems pedantic. There is a difference, at least in language, between mechanical processes and living beings' decision-making. Perhaps instead of "preference", "propensity" would be a more broadly applicable term? [1]https://en.wikipedia.org/wiki/G%C3%B6mb%C3%B6c https://en.wikipedia.org/wiki/G%C3%B6mb%C3%B6c