This sounds a bit like word salad, and your argument is worse than an LLM output. You even contradict yourself at the end. Also I think you need to better understand how much hallucination has been driven down recently and its clear path going forward.
I just asked ChatGPT to give me directions to my favorite shop in Manhattan from Penn Station. It gave me wrong public transit directions about which subway to take. It also gave me wrong walking directions, putting me blocks off course and as to which side of the street I'd find the shop on. The only thing it got right was: "Please note that subway schedules and routes can vary, so it's a good idea to use a navigati…
But why will it get better at accuracy if it isnt trained to be accurate?
What I mean is, the situation of people misusing it should get better over time
Fair. Yes, people will eventually develop best practices through trial & error, and through better features/capabilities of the product/service itself. I just remain extremely cautious and pragmatic during this part of the hype cycle and zone of inflated expectations.
This sounds a bit like word salad, and your argument is worse than an LLM output. You even contradict yourself at the end. Also I think you need to better understand how much hallucination has been driven down recently and its clear path going forward.
I just asked ChatGPT to give me directions to my favorite shop in Manhattan from Penn Station. It gave me wrong public transit directions about which subway to take. It also gave me wrong walking directions, putting me blocks off course and as to which side of the street I'd find the shop on. The only thing it got right was: "Please note that subway schedules and routes can vary, so it's a good idea to use a navigati…
I absolutely agree, and I've been in several similar situations. Yesterday I finally caved and gave it very clear instructions on how I wanted it to build my new IKEA shelf, even though understanding the instructions is most of the work, and guess what? The box is still sitting there, unopened! It hasn't even started. I asked it if it could fight Malenia for me, same thing, more useless machine babble. Repair the slipping rubber on my office chair, incoherent results and not even an attempt to start. Factorise a prime number (I just wanted to break the encryption on some traffic I got a hold of) and it didn't work, again. Just now, I asked ChatGPT what number I was thinking of and it got it so unbelievably wrong... I can't even believe people are using this crap. ChatGPT has very much not come good on the promises OpenAI made about what it could do.
> The 80% statistic refers to the percentage of Fortune 500 companies with registered ChatGPT accounts, as determined by accounts associated with corporate email domains. Yeah... I have no doubt that people at my Fortune 100 company tried it out with their corporate email domains. We have about 80,000 employees, so it seems nearly impossible that somebody wouldn't have tried it. But, since then the policy has come do…
Want about locally running LLM?
I am also currently fighting that fight to internally host something, but right now it is a blanket No for anything AI.
I hereby dub this "Baron-von-Munchausen-as-a-Service." Now you can pay real money for a chatbot to make stuff up about your company and its products. While it is pretty incredible stuff, until or unless they have a veracity bit — some sort of "please don't lie" flag in it, I'd be wary of what it produces. Will ChatGPT offer off-the-cuff pricing, discounts, rebates and refunds that are in line with your actual busines…
I think this is a trap a lot of people fall into: assuming that in ChatGPT solutions that have business value, ChatGPT output is delivered directly to a customer. While that may be true in some cases (e.g. companies that produce cheap blog marketing content should be running scared), it ignores the iceberg of use cases where a ChatGPT tool can simply be one in a set of tools, like Excel is.
Like think of a technical B2B product's technical support engineer who now has a personal assistant trained on the company's full catalog of technical documentation, who can provide instant answers to customer questions that the rep can validate before passing along to the customer on the phone.
Or an overworked public defender who now has an instant paralegal to proof documents or search for and summarize relevant case law that the human lawyer then reviews.
No, what I mean is that it seems as though there is quite a bit of sparseness to the matrix and I was wondering if that can somehow be used to further shrink the model, quantization is another effect (it leaves the shape of the various elements as they are but reduces their bit-depth).
Ah, gotcha! I thought you probably meant something else. I've been wondering this too, and it's something I've been meaning to look at. On a related note it doesn't seem like many local runners are leveraging techniques like PagedAttention yet (see https://vllm.ai/ ) which is inspired by operating system memory paging to reduce memory requirements for LLMs. It's not quite what you mentioned, but it might have a simil…
That's a clever one, I had not seen that yet, thank you.
The hint for me is that the models compress so well, that suggests the information content is much lower than the size of the uncompressed model indicates which is a good reason to investigate which parts of the model are so compressible and why. I haven't looked at the raw data of these models but maybe I'll give it a shot. Sometimes you can learn a lot about the structure (built in or emergent) of data just by staring at the dumps.
It's helpful to think of OpenAI as Microsoft's R&D lab for AI without the political and regulatory burdens that MSR has to abide by. Through that lens, it's really all just the same thing. There is no endgame for OpenAI that doesn't involve being a part of Microsoft.
IIRC it is impossible for OpenAI to become part of Microsoft since the incorporation documents of the for-profit bit of OpenAI prevent anyone from having a majority of the shares (except the non-profit foundation, of course).
Yes, their corporate structure is unprecedented. Very weird and unintuitive.
This sounds a bit like word salad, and your argument is worse than an LLM output. You even contradict yourself at the end. Also I think you need to better understand how much hallucination has been driven down recently and its clear path going forward.
I just asked ChatGPT to give me directions to my favorite shop in Manhattan from Penn Station. It gave me wrong public transit directions about which subway to take. It also gave me wrong walking directions, putting me blocks off course and as to which side of the street I'd find the shop on. The only thing it got right was: "Please note that subway schedules and routes can vary, so it's a good idea to use a navigati…
This is kind of like if a fisherman walking down the street looked up at a steel sky scraper and proclaimed:
"Gee, how can they be making entire buildings out of that?! Why my plastic dinghy could be out at sea for a year and not pick up a lick of rust, meanwhile my steel fishing hooks are rusty if I leave them out on a rainy night!"
Any correlation between this and the sudden disappearance of this repo? https://github.com/microsoft/azurechatgpt Past discussion: https://news.ycombinator.com/item?id=37112741
All activity stopped a couple of weeks ago. It was extremely active and had close to 5 thousand stars/watch events before it was removed/made private. Unfortunately I never got around to indexing the code. You can find the insights at https://devboard.gitsense.com/microsoft/azurechatgpt Full Disclosure: This is my tool
It looks like your account has been using HN primarily (in fact exclusively) for promotion for quite some time. I'm not sure how we didn't notice this before but someone finally complained, and they're right: you can't use HN this way. Note this, from https://news.ycombinator.com/newsguidelines.html: Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity.
Normally we ban accounts that do nothing but promote their own links, but as you've been an HN member for years, I'm not going to ban you, but please do stop doing this! We want people to use HN to read and post things that they personally find intellectually interesting—not just to promote something.
If I go back far enough (a couple hundred comments are so), it's clear that you used to use HN in the intended spirit, so this should be fairly easy to fix.