visarga

179 ms·

2461.
by visarga 4y ago
Still cheaper than a plane, a ship or a power plant, and there are thousands of those.
2462.
by visarga 4y ago
You know what I believe is also a toy model? chatGPT Turbo, you can tell by the speed of generation. And it works quite well, so small size is not an impediment. I expect there will be an open model on the level of chatGPT by the end of the
2463.
by visarga 4y ago
> Whatever grows inside the simulator will never have an understanding that exceeds that of the simulator's maker. Counter example: AlphaGo & AlphaZero, grew inside a Go simulator and surpassed our understanding of the game.
2464.
by visarga 4y ago
Is training a model on second-hand data laundering copyright? Second-hand data is data generated from a model that has been trained on copyrighted content. Let's say I train a diffusion model on ten million images generated by diffusi
2465.
by visarga 4y ago
This technology lowers the entry barrier to almost any field. It empowers self education, creative hacking and growth. Why are you making it sound like a disaster? We can scale our ambitions quickly and absorb the new productivity without l
2466.
by visarga 4y ago
> If AI replaces that level of employee and leaves the seniors, where are the new developers coming from? Why are you assuming this? Juniors with GPT-4 punch above their weight, too.
2467.
by visarga 4y ago
Every 10 years the software paradigms change so much that you practically have to start over. Some skills carry over, but many don't. LLMs will do the same thing - they will force us to change our approaches.
2468.
by visarga 4y ago
I think OP is reacting too much to the implied threat of chatGPT. It's not so good as it seems. If it does the job in 1:10 the time and still makes errors, then it might speed up productivity by a mere 2x because manual review is hard.
2469.
by visarga 4y ago
only works for media and language model data
2470.
by visarga 4y ago
"Open"AI doesn't have time to muck around with hardware design, every month counts.
2471.
by visarga 4y ago
LLMs can be also be used as negotiators. One LLM agent from each party.
2472.
by visarga 4y ago
A LLM trying to trick and manipulate? That's how your multi-million dollar model quickly becomes worthless in the eyes of the public.
2473.
by visarga 4y ago
> But think of hyper relevant ads That depends on ad publishers, right? If they want to sell A, B and C and I am interested in D, then Google's still showing one of A, B or C to me. D doesn't make profit if there is nobody pay
2474.
by visarga 4y ago
> Log all your key thoughts, realizations, decisions and taken steps on paper... Text, bullets, arrows, acronyms, verbs, marks, anything that works. Lists, trees, graphs, outlines ... It’s your L2 cache. This works out nice for using GPT
2475.
by visarga 4y ago
Me too, that's why I don't care how old it is anymore.
2476.
by visarga 4y ago
> but there's nothing a machine puts into this, there's nothing being transmitted That depends on your prompt. You can make it longer if you have more to say. It's a conditional language model, you can select the place it
2477.
by visarga 4y ago
Only if it works flawlessly. If it has subtle bugs you spend about as much time solving the task with GPT as if you didn't use GPT. And it requires deep insight into what might go wrong.
2478.
by visarga 4y ago
Maybe you needed to ask the model to review its own work and especially look for edge cases.
2479.
by visarga 4y ago
Refactoring and cleaning up technical debt might see huge improvements in the future.
2480.
by visarga 4y ago
But LLM black boxes are pretty great at surfacing search leads. They might hallucinate, but they are much more precise semantically than search. LLMs understand when you say something in your own words. In fact one trick is to go to LLM to
2481.
by visarga 4y ago
> Unpredictable Black Boxes Are Terrible Interfaces Like, you know, managing humans?
2482.
by visarga 4y ago
> I cannot see anyone providing an analysis of what that automation will do for those roles Here are two links: > Separate studies of both writers and programmers find 50% increases in productivity with AI, and higher performance and
2483.
by visarga 4y ago
They should use ... you know ... LangChain to write documentations.
2484.
by visarga 4y ago
This one is right in the middle of the action - the plugin market. It is the Android to OpenAI's iOS. Everyone needs a second option.
2485.
by visarga 4y ago
Can you add a library of documents? The model should use search to pad the prompt with relevant demonstration examples before generating the answer. It would be much easier to draw from a known library of text than just using raw GPT.
2486.
by visarga 4y ago
> this looks like a simple usage on top of GPT4, no real IP / innovation You can generate the first draft in the chatGPT-4 window, then ask it to add a section or expand a list and it works pretty well in the chat interface. So it&
2487.
by visarga 4y ago
GPT rolled
2488.
by visarga 4y ago
It's not replacing people, it is doing something else than we do. Even the fact that AI is scalable and doesn't get tired indicates the job replacement idea is wrong. AI will do massive work, something we couldn't have done w
2489.
by visarga 4y ago
When I eat I don't have control or understanding over what is really in my food - microbes, viruses, dirt ... but I still eat without that knowledge. When we use AI we don't understand every weight and activation, but we still get
2490.
by visarga 4y ago
> Every token generated would take as long as it would if you were loading the model from scratch. Wouldn't make sense to keep most of the model in GPU (as much as it fits) and only load the remaining layers on each pass?

More ›