noosphr

41 ms·

931.
by noosphr 2y ago
Humans are not mindless molecules whose only interactions are with the people right next to them.
932.
by noosphr 2y ago
>A lot of their videos look like CGI but apparently the product is real. I bought their robot dog as part of a project to build embodied AI models back in 2022. Their SDK was far more open than anything else on the market and the stock f
933.
by noosphr 2y ago
It scales super exponentially since each person is a member of an arbitrary number of groups and their actions have some non zero impact on each other member of the groups they are a member of, and each individual in each group with a membe
934.
by noosphr 2y ago
This is a perfect example of the limitations of old school social graphs. The number of people you talk to is just the first order effect. What about all the people that people in your company talked to? That has a non zero economic impact
935.
by noosphr 2y ago
Yes, the resulting tensor is incredibly sparse, but still too large to ever be practical for use as anything but the theoretical upper limit on the complexity of a model. The issue is that while you can remove pretty much all possible inter
936.
by noosphr 2y ago
That's only for two way relationships, you need N way relationships, which is N^N.
937.
by noosphr 2y ago
>we do need and should have good fine-grain computational models of complex systems such [...] the global economy. Many years ago when 'social graphs' were still a hot area to do research in I started building a simulation of t
938.
by noosphr 2y ago
People not in the field have no idea just how distorted the market is right now. I was working at a startup doing end to end training for modified BERT architectures and everything from buying a GPU - basically impossible right now, we ende
939.
by noosphr 2y ago
NVidia only has so much silicon area it can get from its partners. Right now AI datacenter silicon is worth much more than its weight in gold, but consumer grade silicon is worth a fraction of that. I predict we are pretty close to the deat
940.
by noosphr 2y ago
This is easily solvable by using _plain English_ as the interop layer. We have more or less solved natural language input with open source llms that are about as capable as llama3-70b. Searching the docs of the next tool to be called is all
941.
by noosphr 2y ago
https://xkcd.com/882/
942.
by noosphr 2y ago
The AI company I've been working at ran out of money last week so I'm taking a month long break. I've been playing around with defining a standard that is easy to implement for serializing tabular data using the ASCII delimit
943.
by noosphr 2y ago
This is from 2018, anyone in the field know if it's still state of the art or a historic curiosity? I know that we've started using euv since then which seems like it would change things.
944.
by noosphr 2y ago
I was doing something very similar to telelumen but given the variation in LED spectrum you could get off aliexpress in 2020 I could never hope to match their quality without tuning each led separately. Looking at the spectrum graphs for yo
945.
by noosphr 2y ago
Are you guys looking at fabbing your own LED dies? The actual spectrum of commercial LEDs is all over the place when you start measuring it it with a spectrometer, even when they supposedly have a high CRI. Especially if you want some tempe
946.
by noosphr 2y ago
The difference is that the coma and newline characters are much more common in text than 0x1F and 0x1E, which if you restrict your data to alphanumeric characters (which you really should) will never appear anywhere else.
947.
by noosphr 2y ago
> but these are the right first steps. What are you talking about? The concept of a hacker space was exported to the US from the CCC which started in 1981 in _West Berlin_. That idea mutated a bunch of times to become more palatable to t
948.
by noosphr 2y ago
Depends on the country. In the US it's ~10% of the population. Completely doable, especially since the participation rate in the economy is still lower than at any point in the last 50 years (minus covid). If that seems to hard I guess
949.
by noosphr 2y ago
A new technology has made old modes of thought obsolete. A generation which grew up fetishising those things when they were young is now upset they are no longer valuable or meaningful. Or to quote Socrates on the invention of writing: >
950.
by noosphr 2y ago
>Try doing that as a teacher with a classroom of 30+ students. I believe that OP is saying that's exactly the problem.
951.
by noosphr 2y ago
>Of course the model may hallucinate, but in this case it takes a few clicks in the dev tools to verify that this is not the case. I don't know - or care to figure out - how OpenAI does their tool calling in this specific case. But
952.
by noosphr 2y ago
>The tool name is not relevant. It isn't the actual name, they use an obfuscated name. >EDIT: And googling the tool name I see it's already been widely discussed on twitter and elsewhere I am so confused by this thread.
953.
by noosphr 2y ago
Posts like this are terrifying to me. I spend my days coding these tools thinking that everyone using them understands their glaring limitations. Then I see people post stuff like this confidently and I'm taken back to 2005 and arguing
954.
by noosphr 2y ago
>You can ask 4o about this yourself. Here's what it said to me: >"So while I’m deeply multimodal in cognition (understanding and coordinating text + image), image generation is handled by a linked latent diffusion model, not
955.
by noosphr 2y ago
Then use it to mine monero or similar. The idea that you should pay for content shouldn't be an insane pipedream. It should be the default on the internet. Maybe then we wouldn't be in the situation where getting new users is an e
956.
by noosphr 2y ago
But that's OPs point. You need to care about those hyper parameters. Good on them for finding them in smaller models, but they still had to find them, because without those hyper parameters pre-training is at best unstable and at worst
957.
by noosphr 2y ago
>How much would it cost to fine-tune an AI to recognize fish and do this automatically instead of crowdsourcing? $0 basically. You can fine tune a YOLO model on a work station in an hour or so.
958.
by noosphr 2y ago
There are basically no people who have done pre-training of LLMs. What you're describing is absolutely the case, but everyone not in a top 100 AI lab would have never had the displeasure of having to deal with that. That said I strongl
959.
by noosphr 2y ago
>what i know is that in my particle simulations and multithreaded compilation tasks it is easily 2-10x faster. If Moore's law still worked like it did between 1975 and 2005 you'd be getting a 16x performance boost in _single_ t
960.
by noosphr 2y ago
>Of course as the Moore law refuses to die, we'll continue seeing 1.5-2x or so every year, but that's far from 10x. This is another one of those anec-data throw away sentences that take thousands of words to disprove - with a l

More ›