jxy

14 ms·

121.
by jxy 3y ago
Glory to Mankind
122.
by jxy 3y ago
I don't understand the argument. Isn't the whole point of PGP establishing some kind of chain of trust? If pypi.org has it's public key, it could sign a few major distributors's keys, and for smaller/individual pack
123.
by jxy 3y ago
Bloom models are hopelessly under-trained. This one is worse than a 13B Vicuna.
124.
by jxy 3y ago
Speaking of which, HBO now has the perfect chance of riding this LLM hype train with a new season of Westworld, only if they hadn't cancelled it.
125.
by jxy 3y ago
True. I can't believe how simple OpenAI made their API to be. They really made it trivial to start using their API. Comparing that to Google's, words fail me to even describe the frustration with GCP just to start testing it'
126.
by jxy 3y ago
It should be emphasized that Bing never processes the link in real time. It really just access some Microsoft database when you give it a link. Sometimes it's frustrating to realize that Bing hallucinated the whole article based on the
127.
by jxy 3y ago
This seems to be a setup of an extremely interesting experiment. We could get a bunch of different language models and actual humans together and prompt them to write under these different scenarios.
128.
by jxy 3y ago
They must hate lisp so much that they opt to use {{}} instead.
129.
by jxy 3y ago
I guess I only know transformers and how BERT or GPT works, as there would be a limit in the context length. With GPT, you can certainly generate infinite amount of tokens but the previous tokens outside of the maximum context length would
130.
by jxy 3y ago
There is a limit of maxOutputTokens, 1024! Is this the true capability of PaLM 2? However I couldn't find anything about the context length of their model anywhere. And the API didn't tell me how long the prompt could be.
131.
by jxy 3y ago
So how do we actually try out the PaLM 2? The links in their press release just link to their other press release, and if I google "PaLM API" it just gives me more press release, but I just couldn't find the actual document f
132.
by jxy 3y ago
They should have just used standard R7RS small. Creating a variant only serves more confusion.
133.
by jxy 3y ago
I really like the formatting of the document pages there, monospaced font with nice graphics, very legible table of contents, and links and back links with see also and back to. If I'm not using Dyalog APL, are there other tools that
134.
by jxy 3y ago
The problem is now they know your TOTP secrets, they are only one password away from pretending to be you. And actually, they serve you emails, so password is moot for most of the sites today.
135.
by jxy 3y ago
What LLM are you using? How do you save and load the persistent state? And how do you deal with the limited context window of the LLM you are using?
136.
by jxy 3y ago
Well, just like Printing press had destroyed history, religion, culture, literature, and etc. Radio had destroyed music industry. Cassette player had destroyed music industry the second time. Internet had destroyed music industry the third
137.
by jxy 3y ago
Many car insurances in the US are giving people incentives to install a tracking app in people's phones. I won't be surprised when they start handing out dash cams with audio recordings.
138.
by jxy 3y ago
In terms of pickled files, we may ask them to release the models in safetensors. That would be a big improvement. There's not much we can do about dependencies on pytorch or other python libraries. Perhaps people can make more independ
139.
by jxy 3y ago
If I understand the image processing part [1] of the code correctly, each image is scaled down (max 448x224) and only uses <= 512 tokens. 14x14 pixels is one input token. [1] https://github.com/haotian-liu/LLaVA&#x2F
140.
by jxy 3y ago
This opens a huge possibilities. It's likely we could simply plug in stable diffusion using a linear layer. As well as whisper and some TTS. Getting a back to back mixed image/sound/text engine running on a laptop. I wonder i
141.
by jxy 4y ago
Right. The nature trained all the nervous systems with genetic algorithms for millennia. Individuals rather fine tuned their own with natural noise and dropoffs and natural feedback. Fine tuning our brains are rather expensive.
142.
by jxy 4y ago
I guess it's joke, but I had to try. GPT-4's interpretation of the prompt is "on point": class ScientistSkillClassifier: def __init__(self): self.skill_domains = { "biology
143.
by jxy 4y ago
GPT-4 answered correctly and explained the chemical reaction.
144.
by jxy 4y ago
Even the 4bit quantized llama 13B tuned with rlhf-lora on alpaca dataset got it right. sampling parameters: temp = 0.100000, top_k = 40, top_p = 0.750000, repeat_last_n = 64, repeat_penalty = 1.000000 Below is an instruction th
145.
by jxy 4y ago
I was very curious and checked with this string, "eeeeeeggeeeee", which should be tokenized as [eeee][ee][g][ge][eeee]. Both GPT3.5 and GPT4 gave me "egg", which is a single token.
146.
by jxy 4y ago
The decades old poison of Micro$oft: Embrace, extend, and extinguish. On the hardware level, it feels like computing just has gone back to the era of mainframes. Hopefully the era of PC soon comes and everybody can start running their own
147.
by jxy 4y ago
Change your browser's user agent to Edge.
148.
by jxy 4y ago
To use "the good Bing", you only need to change the user agent of your favorite browser to Edge.
149.
by jxy 4y ago
Maybe we just need a better attribution, so people can decide on the level of objective confidence. If you live in the US, it is already difficult to avoid seeing and listening to real people lie publicly, which may take journalists a long
150.
by jxy 4y ago
> Note, for all the seemingly sophisticated thought and language, the moral indifference born of unintelligence. I would think that ChatGPT's response about morality is a typical canned response written by OpenAI. text-davinci-003 c

More ›