slashdave

9 ms·

151.
by slashdave 5mo ago
What happens when they run out of letters in the alphabet?
152.
by slashdave 5mo ago
Dunno. Isn't OpenAI supposed to release a new version of their model within 30 minutes? Maybe things are actually quieting down.
153.
by slashdave 5mo ago
Are we supposed to have two cars?
154.
by slashdave 5mo ago
What? No, that is not what quantum computers do
155.
by slashdave 5mo ago
I understand what you are saying, but there are many true believers out there
156.
by slashdave 5mo ago
RL is more than facts. Synthetic feedback is an obvious approach. Does the model suggest code that compiles and performs well?
157.
by slashdave 5mo ago
I think you are assuming training from scratch, which I doubt is happening here. Fine-tuning and RL, especially based on synthetic feedback (coding skill, in particular) can be ongoing and is where these models obtain truly useful abilities
158.
by slashdave 5mo ago
It might be possible that in a few years someone will be able to engineer a reasonably priced machine to run today's frontier models (hint, your price is an order of magnitude off). However, they won't be able to run the frontier
159.
by slashdave 5mo ago
> We cannot condone naïve enthusiasms, nor fuel unfounded fears
160.
by slashdave 5mo ago
Genetic manipulation of humans, both indirect and direct
161.
by slashdave 5mo ago
> It is a golden era for buckets and buckets of slop, and a dark age for gems of quality. I mean, this has been the trend for decades really, before LLMs were a thing. The incentive is skewed toward quantity rather than quality. The new
162.
by slashdave 5mo ago
Define "arbitrary". Without RAG, they screw up basic algebra.
163.
by slashdave 5mo ago
LLMs are statistical models by construction, so depending on how liberal you want to be with terminology, "interpolate" is not so bad. Might make a statistician upset.
164.
by slashdave 5mo ago
You have to define what you mean by "interpolate". The mechanisms that LLM use are not mysterious, and they are not the same as used by humans.
165.
by slashdave 5mo ago
I am not entirely sure you understand how expensive it is to train these models
166.
by slashdave 5mo ago
So... do you see a problem with regulating skynet to not kill us all?
167.
by slashdave 5mo ago
Yep. You can also put your key on a usb drive that can be read on boot. Just choose a good quality one....
168.
by slashdave 5mo ago
We are discussing how rapid development has been, and now you want to freeze your model in silicon?
169.
by slashdave 5mo ago
Print it on a piece of paper and put it in a lock box.
170.
by slashdave 5mo ago
> within a few years we will be running local models as good as today’s frontier models I seriously doubt it. Scaling is already strained (don't buy into the "exponential" hype). And, in any case, the competition will be a
171.
by slashdave 5mo ago
> Every software developer knows that you can’t make projects go faster just by typing faster. You know, typing fast and accurately is kind of important. The new speed skill that developers now need is speed reading. LLMs just make copio
172.
by slashdave 5mo ago
Says the engineer and not the designer
173.
by slashdave 5mo ago
Well, in the good old days, when we have refereed journals, it would be part of the publishing process.
174.
by slashdave 5mo ago
Tokens are not a substitute for a numerical measurement. Ask a LLM how much time has passed. Watch it hallucinate wildly. Has anyone noticed that Opus has trouble building ascii diagrams (often leaves out spaces so lines are misaligned)?
175.
by slashdave 5mo ago
I mean, sure, reduced complexity is great, but... what about performance?
176.
by slashdave 5mo ago
Because they (Google) are so removed from manufacturing
177.
by slashdave 5mo ago
Some of us actually want to get work done
178.
by slashdave 5mo ago
You forgotten the important part: permissions
179.
by slashdave 5mo ago
They are also surprising good at finding bugs that humans often miss
180.
by slashdave 5mo ago
> like AGILE and SCRUM Yeah, likely > development patterns that are good practices. Wait, now you lost me

More ›