jmiskovic

9 ms·

61.
by jmiskovic 3y ago
All of these should have close ties with OpenAI, but no one from inside confirmed it as far as I know. https://twitter.com/swyx/status/1671272883379908608 https://twitter.com/soumithchintala/s
62.
by jmiskovic 3y ago
It's been confirmed by multiple (unofficial) sources that GPT-4 is 8 models, each 220B parameters. Another rumor is GPT-4 being 16x111B models. There's a quite fresh and active project replicating something similar with herd of ll
63.
by jmiskovic 3y ago
Open source is both a colloquial term for available/modifiable/distributable code (like Llama2), and a strict OSI-approved list of licenses. I'd say opensource-ish is a great fit here. Edit: this is in fact fairly interesting
64.
by jmiskovic 3y ago
Nope. It's limited by 700m monthly-active users at the time Llama2 was released, a weird catch clause for a handful of Meta competitor companies. The license doesn't satisfy OSS requirements, but it is quite reasonable. https:&#x
65.
by jmiskovic 3y ago
Probing the conscious and subconscious mind, a novel mind-boggling 3D-to-2D visualization, working interactive demo, nice hand-drawn diagrams - this is my HN nugget of the day. Kudos!
66.
by jmiskovic 3y ago
I use ffmpeg often because it's so powerful, but its api cannot fit in my head. It should have a LLM frontend.
67.
by jmiskovic 3y ago
There is no fully built solution, only bits and pieces. I noticed that llama outputs tend to degrade with amount of text, the text becomes too repetitive and focused, and you have to raise the temperature to break the model out of loops.
68.
by jmiskovic 3y ago
It's actually 175B. https://arxiv.org/pdf/2005.14165.pdf
69.
by jmiskovic 3y ago
That's rather strange. The graphics part is lightweight (pre-rendering the background and then drawing few shapes), but if you could shrink the browser to very small dimensions and test we could eliminate this one. The audio part is bi
70.
by jmiskovic 3y ago
The 70B Llama2 model ties in with 173B ChatGPT-0301 model. The GPT-4 still stands unchallenged.
71.
by jmiskovic 3y ago
Then why not read their paper? "The largest Llama 2-Chat model is competitive with ChatGPT. Llama 2-Chat 70B model has a win rate of 36% and a tie rate of 31.5% relative to ChatGPT."
72.
by jmiskovic 3y ago
I made a fork with few more features, it might even work on your phone browser: https://jmiskovic.github.io/voicebox
73.
by jmiskovic 3y ago
Another World is on a whole other level. SCUMM is from '89 and the NDS came out in 2004. Another World game came out in 1991, and because it used the VM it could be back-ported to Apple IIGS (1986), the computer that's 5 years old
74.
by jmiskovic 3y ago
DRM is bad, most of us agree on this. Also, copyright laws were written for bygone era where things didn't move so fast. I think it's the case of the "lawful" not being aligned with "ethical". But the tragedy o
75.
by jmiskovic 3y ago
The Duke is another chess variant I'm fascinated by. It's still on rectangular grid, with more varied rules for each piece movement and attack patterns. What I find interesting is that these rules are inscribed on game pieces them
76.
by jmiskovic 3y ago
Don't they aggressively fight this? I think all your accounts can get banned if you use multiple accounts. I know they demand a working phone number and other identification data. Not saying there are no multiple accounts, just that it
77.
by jmiskovic 3y ago
"Very tedious without it" doesn't sound like just 5% improvement? I've started developing in a new language and I can hardly do any work without the LLM assistance, the friction is just too high. Even when auto-competiti
78.
by jmiskovic 3y ago
Those are not really useful for getting the feel of federated communities. Check out the lemmy browser & search around for your hobbies and interests: [0] https://browse.feddit.de/
79.
by jmiskovic 3y ago
Weirdly specific off-topic discussion. I tried the same thing two days ago and the mouse scrolling worked well - I could scroll through the system menus using the wheel on BT mouse. The termux brings the power of linux to Quest. You can hav
80.
by jmiskovic 3y ago
A common courtesy would be to check for naming collisions in voice-generating domain. https://github.com/jmiskovic/voicebox If I ever start selling scrapbooks for collecting human faces I'll be returning the favor
81.
by jmiskovic 3y ago
Nope. The model itself is GPL-3, legally it doesn't matter if you use Meta implementation or llama.cpp. Weights being subject to copyright was never tested in court. Businesses have wisely steered away from using leaked weights for com
82.
by jmiskovic 3y ago
The GPT4 cognitive leaps and breadth of knowledge seem impossible to replicate with any number of 13B models. 13B x10 might be the only feasible business model for your company, but that alone doesn't make it a superior option. Can you
83.
by jmiskovic 3y ago
There's a LocalLLaMA Lemmy instance at https://sh.itjust.works/c/localllama
84.
by jmiskovic 3y ago
First time I heard of it. I like how simple the core concept is explained, but I feel it would box me into categories when I like tags more (categorizing items in multiple orthogonal domains). OTOH maybe well thought-out categories would br
85.
by jmiskovic 3y ago
You can run it on RPi or any old hardware, only limited by the RAM and your patience. It is a lean code base easy to get up and running, and designed to be interfaced from any app without sacrificing performance. They are also innovating (o
86.
by jmiskovic 3y ago
As a VR hobbyist dev I'm divided on how I feel about the device. Price thing is already well discussed, for me it's a tool worth investigating if I should invest time in the ecosystem. Without a doubt, the Vision Pro hardware is a
87.
by jmiskovic 3y ago
Demo isn't really connected to the model, no way to test it out. Last commit was two weeks ago. Did anyone reproduce the outlined steps? If this thing can be stronger than GPT-3.5 at 7B parameters (which I find unlikely), it would make
88.
by jmiskovic 3y ago
You think it is a good idea to track how eye subconsciously moves, and to trust that data with Meta? Don't mind if I skip that tech lane.
89.
by jmiskovic 3y ago
It's ongoing effort. At first they scrapped ShareGPT and used it to train the Alpaca model. After that others have pruned the dataset to remove examples where ChatGPT refused to answer. These datasets and resulting models are called &q
90.
by jmiskovic 3y ago
Correcting the correction of a correction: The chat interface is no longer using GPT-3.5, since May 3rd. What is the default model called? I have no idea. The release notes page [1] calls it ChatGPT model, the bottom of chat interface calls

More ›