solidasparagus

9 ms·

91.
by solidasparagus 2y ago
That, as I understand it, is not a valid chain of logic. Requiring fewer data points does not inherently indicate that the underlying mechanism (autogressive sequential generation, not the transformer which is just an architecture) is diffe
92.
by solidasparagus 2y ago
Tell me more?
93.
by solidasparagus 2y ago
An article this long needs a table of contents IMO.
94.
by solidasparagus 2y ago
> AI needs to see thousands or millions of images of a cat before they reliably can identify one. Not if they inherit from a previous generation of AI. But even if they did, a different training speed does not imply a different capabilit
95.
by solidasparagus 2y ago
It is how it works if you are replying to someone who claims "If you really think about what an LLM is you would think there is no way that leads to general purpose AI". The counter example is human beings are considered general p
96.
by solidasparagus 2y ago
The vast majority of people use a model built by someone else through a high-level abstraction and then make broad claims about what the technology will be capable of. Then you have the people who are leveraging the technology to train mode
97.
by solidasparagus 2y ago
> If you really think about what an LLM is you would think there is no way that leads to general purpose AI It is an autoregressive sequence predictor/generator. Explain to me how humans are fundamentally different
98.
by solidasparagus 2y ago
What do you mean for the assistant to be yours? There is a core model somewhere that needs to be trained and the cost/technical requirements are prohibitive for that to be done by each individual. If the model is local, is that enough?
99.
by solidasparagus 2y ago
NVIDIA has fundamental hardware and software advantages that will take years and years to catch up on. The general purpose part of GPGPU is no joke, just ask the people trying to build competing hardware and an alternative to the CUDA stack
100.
by solidasparagus 2y ago
I feel like you are complaining about the first car. Yes, it sucks and, god, I'm not sure it's good for the world. But zoom out a bit and it seems fairly clear to me that this is what the future is going to look like and this is w
101.
by solidasparagus 2y ago
FYI for anyone else skimming - requires you to do a coding problem for your application to even be read.
102.
by solidasparagus 2y ago
That's very true. Although it is feasible if you are well-resourced and make the investment to own the toolchain end-to-end. Serving costs can be quite low (relatively speaking) if you control everything. And you have to pick the corre
103.
by solidasparagus 2y ago
People will disagree, but my problem with retrieval is that every technique that is popular uses one-hop thinking - you retrieve information that is directly related to the prompt using old-school techniques (even though the embeddings are
104.
by solidasparagus 2y ago
That leaves a market gap and then a better company can fill it, unlike centralized economies. Is that happening? I would say yes, at least in some cases. High-quality, direct-to-consumer brands are great. I rely on purchasing guides for man
105.
by solidasparagus 2y ago
Huh. Well that's embarrassing. I guess I missed it when I lost interest in the caching section and jumped straight to Evaluation and Monitoring.
106.
by solidasparagus 2y ago
I took at look at Magic earlier today and it didn't work at all for me, sorry to say. After the example prompt, I tried to learn about a table and it generated bad SQL (correct query to pull a row, but with limit 0). I asked it to show
107.
by solidasparagus 2y ago
Continuous retraining and deployment maybe? But I'm actually not anti-RAG (although I think it is overrated because the retrieval problem is still handled extremely naively), I just think that fine-tuning should also be in your toolk
108.
by solidasparagus 2y ago
Hmm, I don't want to talk specifics about my experience, but maybe check out some of the case studies on Surge's website - https://www.surgehq.ai/ (about halfway down the page). > Do they ask PhD to explain roo
109.
by solidasparagus 2y ago
What sort of leak? I've seen data labeler/generation teams hired. I've never heard anyone describe the existence of these teams as a secret. No one hides the existence of Scale AI. People talk about which providers are better
110.
by solidasparagus 2y ago
> we <3 RAG, and GNNs for improving your KG is fascinating Oh man I am so torn between this being a fantastic idea and this being "building a better slide-rule in the age of the computer". dspy is definitely a project I want
111.
by solidasparagus 2y ago
100 percent accurate will never happen, nor does it need to. But think about the intelligence of an average human. Can we beat that? At least along some collection of concrete axes, enough to create a form of intelligence that can rightfull
112.
by solidasparagus 2y ago
Understanding is a loaded term. But large transformers seem pretty good at learning from datasets (of more or less any modality) to they extent that they can create useful new datapoints and allow you to work with existing datapoints in use
113.
by solidasparagus 2y ago
See if the article said this, I would have agreed - fine-tuning is a tool and it should be used thoughtfully. Although I personally believe that in this funding climate it makes sense to make data collection and model training a core capabi
114.
by solidasparagus 2y ago
> A well-trained 8B model will already be over-saturated with information from the start Any evidence of that that I can look at? This doesn't match what I've seen nor have I heard this from the world-class researchers I have w
115.
by solidasparagus 2y ago
> Does no one remember why expert systems fell apart? Many of the current generation of AI experts mostly either did not pay great attention to the history of AI or they believe this time is completely different. They would do well to sp
116.
by solidasparagus 2y ago
> I don't consider a small 8B model to be worth fine-tuning. Going to have to disagree with you on that one. A modern 8B model that has been trained on enough tokens is ridiculously powerful.
117.
by solidasparagus 2y ago
How much of that is just the flood of traditional engineers into the space and the fact that collecting data and then fine-tuning models is orders of magnitude more complex than just throwing in RAG? I suspect a huge amount of RAG's po
118.
by solidasparagus 2y ago
It is not unfeasible. It is absolutely realistic to do distributed finetuning of an 8B text model on previous generation hardware. You can add finetuning to your set of options for about the cost of one FTE - up to you whether that tradeoff
119.
by solidasparagus 2y ago
Thanks, I'll read those more fully. But "knowledge injection" is still pretty narrow to me. Here's an example of a very simple but extremely valuable usecase - taking a model that was trained on language+code and finetun
120.
by solidasparagus 2y ago
No offense, but I'd love to see what they've successfully built using LLMs before taking their advice too seriously. The idea that fine-tuning isn't even a consideration (perhaps even something they think is absolutely incorr

More ›