swalsh

14 ms·

181.
by swalsh 2y ago
Yeah, this one needs to be resolved. It really is confusing since they're in the same space.
182.
by swalsh 2y ago
GPT-4 degraded significantly, but you probably have some rosey glasses on. Sonnet is signifcantly better.
183.
by swalsh 2y ago
And with the increasing parameter size, the main winner will be Nvidia. Frankly I just don't understand the economics of training a foundation model. I'd rather own an airline. At least I can get a few years out of the capital i
184.
by swalsh 2y ago
vLLM is nice because you can have multiple streams. I believe the tech is called batched attention, but working from memory so I might be wrong. When scaling up inference, this feature is super key. This is the difference between 70 toke
185.
by swalsh 2y ago
Probably the best play of the cycle if you're not Nvidia
186.
by swalsh 2y ago
I have a theory that the days of established businesses that don't know tech is dwindling. A lot of companies which has adopted tech has started building a small foundation of talent internally. I think you're seeing this trend a
187.
by swalsh 2y ago
Going to be vague, but I'm using it to scale out human processes in ways I couldn't using humans (because they cost too much) or regular code (because it's unstructured). Early results are promising, we've found a bunch
188.
by swalsh 2y ago
I've worked with hortonworks, cloudera, and databricks. It's no surprise at all that databricks is killing the competition. Those other companies products were embarrassingly terrible. Not stable, slow, and worst of all, I had
189.
by swalsh 2y ago
I'm not sure I'm exactly at the edge of things, but I have 2 companies trying to setup regular meetings with me to be a beta customer. Both have promised I can help define a new product, but when I list my real problems... they a
190.
by swalsh 2y ago
Why is this article from 2021 being posted now?
191.
by swalsh 2y ago
Last time I emailed them, I think they said they're trying to release pro in July... so should be coming soon.
192.
by swalsh 2y ago
Unsloth is a great tool, super fast.
193.
by swalsh 2y ago
LoRA is perfect for that. Where it fails is when you want to teach a new domain.
194.
by swalsh 2y ago
After about an hour of using this new model.... just WOW this combined with the new artificats feature, i've never had this level of productivity. It's like Star Trek holodeck levels. I'm not looking at code, i'm descr
195.
by swalsh 2y ago
Anthropic has been killing it. I subscribe to both chatgpt pro and claude, but I spend probably 90% of my time using Claude. I usually only go back to open ai when I want another model to evaluate or modify the results.
196.
by swalsh 2y ago
Running into this same problem, the tech is progressing WAY faster than the org can transform to use the tech. I'm betting we end up transforming by aquiring a company that completed the transformation before us, and letting them run
197.
by swalsh 2y ago
Along the same lines, phi-3 is kind of a sign of what you can do if you focus only on high quality data. It seems like while yes, quantity is very important, quality almot matters just as much.
198.
by swalsh 2y ago
I agree, all 3 were great viable answers to my question. But Claude and GPT-4 felt a lot more like a regurgitation of the same suggestions people have proposed for years (ie, their training material). The GPT2, while similar felt more lik
199.
by swalsh 2y ago
I'm impressed. I gave the same prompt to opus, gpt-4, and this model. I'm very impressed with the quality. I feel like it addresses my ask better than the other 2 models. GPT2-Chatbot: https://pastebin.com/vpYvT
200.
by swalsh 2y ago
Putting kids into a technology induced coma can be useful in specific situations... such as on an airplane. But it should generally be a last resort, and avoided. Many parents use it too quickly.
201.
by swalsh 2y ago
Very troubling. In my opinion social media in general is like smoking. It's a generally negative thing that's addictive... but also pleasurable. Adults should be allowed to made decisions on what they do to themselves, but chi
202.
by swalsh 2y ago
Sorry, i used some jargon most people don't know. Prepay the way I meant it is not referring to prior auths, but rather pre-adjudication. Not everything can go through a pre-pay process, but I believe most claims can.
203.
by swalsh 2y ago
I think as the industry moves from a post-pay model, to a pre-pay/real-time model, this antagonistic relationship between providers and payers should slowly improve. Definately not a silver bullet though.
204.
by swalsh 2y ago
This is really cool. As someone who knows enough to get something working, but nothing more, I've always dreamed of something like you're describing.
205.
by swalsh 2y ago
Parameter count seems to only matter for range of skills, but these smaller models can be tuned to be more than competitive with far larger models. I suspect the future is going to be owned by lots of smaller more specific models, possibly
206.
by swalsh 2y ago
Why is Meta doing it though? This is an astronomical investment. What do they gain from it?
207.
by swalsh 2y ago
Gamers have a TON of really good really affordable options. But you kind of need 24gb min unless you're using heavy quantization. So 3090 and 4090's are what local llm people are building with (mostly 3090's as you can get
208.
by swalsh 3y ago
The being too precise reduces accuracy example makes sense to me based on my crude understanding on how these things work. If you pass in a whole list of states, you're kind of making the vectors for every state light up. If you just
209.
by swalsh 3y ago
None of these "songs" have any emotion.. AI music just doesn't make me "feel" anything yet.
210.
by swalsh 3y ago
I'm not using the API. Both are $20/month subscriptions.

More ›