kromem

22 ms·

301.
by kromem 3y ago
I'm curious when we'll start to see open access multimodal models being released. The advancement in text only models has been amazing, but a lot of the 'emergent' behavior in GPT-4 may be because of multimodal training
302.
by kromem 3y ago
Actually, if you read Sy Damile's (a former General Counsel of the US Copyright office) testimony to Congress (which is really worth a read), he makes a very good point that the smaller the training data set the less likely that the
303.
by kromem 3y ago
Because they need the ocean, even if they don't need any individual drop within it. Relative to the rest of the data, NYT's is not meaningfully impactful. But they need the aggregate amount of data, which is why they need training
304.
by kromem 3y ago
I've been fairly persuaded that quantum mechanics are the side effects of simulating a continuous universe (how ours appears at macro scales). So even though 99% of the universe models one from 'real' (mathematically) origins
305.
by kromem 3y ago
Which was in turn strongly informed by the Epicurean version by way of Lucretius. Both of which really fail to think through the concept in light of their views of continued evolution. If humanity is but a stepping stone to something greate
306.
by kromem 3y ago
Yes!!! This has been one of the most interesting things I researched over the pandemic and it's AMAZING. So what a lot of people might not really know is that many modern ideas are older than we think, specifically compiled into a poem
307.
by kromem 3y ago
Link to the PDF of the research itself for those interested: https://www.journalofscience.net/showpdf/MjY4a2FsYWkxNDc4NTI...
308.
by kromem 3y ago
They also seem to pass the mirror test: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9881685/#:~:tex... .
309.
by kromem 3y ago
Others provided good advice about technically doing it. A note in terms of the project goals: Make sure to remember when interpreting your results that your findings will only apply to similarly sized models as what you trained. So you&#x27
310.
by kromem 3y ago
Perhaps the most important point in the piece, and one that can't be repeated enough or understood enough as we head into what 2024 has in store: > And then, do LLMs have some reasoning abilities, or is it all a bluff? Perhaps at ti
311.
by kromem 3y ago
Weird, as I see it as almost the flip side of the coin from the author. I was just today discussing my playing with a resolution to - by year end - go back to a flip phone. But the only way I see this as feasible in a modern world, to disco
312.
by kromem 3y ago
At this point algo trading is a race to exploit the other algo created patterns, and is less predicting inherent market inefficiencies as much as fellow algo trading inefficiencies. Self-fulfilling prophecy is a thing, even when the prophec
313.
by kromem 3y ago
While true, it's quite an underappreciated factor and rarely considered seriously by people. You are correct that the human ability to manifest our imagination is remarkable. We dream of flying like a bird, eventually we build machines
314.
by kromem 3y ago
Training is probably going to turn out to be fair use as the suits settle: https://www.eff.org/deeplinks/2023/04/how-we-think-about-cop... It's the usage and not the training that needs to be policed, an
315.
by kromem 3y ago
Star Citizen could release and immediately be pirated and never sell a single copy from release onwards and likely still have ended up profitable. Maybe the business models around creation need to be revisited such that interested parties p
316.
by kromem 3y ago
Here's one of the senior legal peeps at the EFF who has litigated IP cases talking about the issue: https://www.eff.org/deeplinks/2023/04/how-we-think-about-cop... It's not as clear cut as you thin
317.
by kromem 3y ago
Dreidel games don't last very long without valued 'currency' like chocolate. You're correct there's no intrinsic motivation, which is why the game has extrinsic motivators.
318.
by kromem 3y ago
You said "I'm pretty sure you mean Dirk Englund's photonic matrix multiplication." No, I don't mean his photonic matrix multiplication from 2022. The issue seems more to be with your over specificity leading to ambi
319.
by kromem 3y ago
Conversely, in an age before TV or Internet, the longer the game could keep kids occupied the better, no? We might even imagine that anthropologically given the role of keeping kids occupied there were other games that were more fun but of
320.
by kromem 3y ago
Then we should be focused on policing the usage of the model, not the training of it. That's the point at which infringement occurs in your example. It's not the memorizing that's the infringement, it's the reproduct
321.
by kromem 3y ago
"Let's ban something capable of diagnosing medical conditions and letting coma patients to communicate with an EEG because it learned the relationships between words from a giant data set of scraped data and is owned by a company&
322.
by kromem 3y ago
That's not the case, as they aren't trying to get a ruling on the forced reproduction by prompt as infringement, but rather to get a ruling that training is infringement. I fully agree with the perspective that infringement in us
323.
by kromem 3y ago
It depends on how much is reproducible and what the use is. If only small patches of the original image can be reproduced then it becomes much more murky.
324.
by kromem 3y ago
This isn't an issue with training, it's an issue with usage. Production open access LLMs do probably need a front-end filter with a fine tuned RAG model that identifies and prevents spitting out copyrighted material. I fully sup
325.
by kromem 3y ago
No, that was 2022. I'm talking about this one: https://dl.acm.org/doi/10.1145/3603269.3604821
326.
by kromem 3y ago
It's not really any different from the current state of models, it's just that moving the same constraints to a different hardware foundation opens up significant performance gains.
327.
by kromem 3y ago
General purpose QC isn't going to happen all that soon, and may always be faced with scaling issues. But we will see QC in a very specific application before the decade is out. It's kind of a perfect marriage between ML and QC.
328.
by kromem 3y ago
I suspect this is Llama-2-chat and not the base model. It's very attuned to safety fine tuning, and it isn't having issues with completion in a chat format. They probably weren't specific enough in mentioning what the model i
329.
by kromem 3y ago
It seems like it must be using Llama-2-chat, which has had 'safety' training. To test which underlying model I asked it what a good sexy message for my girlfriend for Valentine's Day would be, and it lectured me about objecti
330.
by kromem 3y ago
I found the multiple models getting this one correct really interesting: As a result of an accident, Abdul lost sight in his right eye. To judge the distance of vehicles when he is driving, Abdul is able to rely on cues of A. I only

More ›