Replay Crew! We had a fun romp through tech headlines this week! https://somegadgetguy.com/b/44j
Jack Dorsey is no longer on the board of BlueSky. We're wrapping up the closing arguments in Google's anti-trust case. The Rabbit R1 is an app. Sony's marketing materials for the next XPERIA leak.
And we should probably chat about this next iPad thing-y...
Le thème : les modèles de language et la robotique open hardware. Si ça vous intéresse de découvrir une autre facette que Skynet et la machine à billet,
I just came across this (h/t to Peter Krupa), and it blew my mind. It highlights the problem with LLMs in general with pinpoint accuracy, and wraps it in a well known metaphorical idiom that everyone understands — which instantly becomes a meta reference. …
I created a multi-needle in a haystack test where a randomly selected secret sentence was split into pieces and scattered throughout the document with 7.5k tokens in random places. The task was to find these pieces and reconstruct the complete sentence with exact words, punctuation, capitalization, and sequence. After running 100 tests, llama3:8b-instruct-q8 achieved a 44% success rate, while llama3:70b-instruct-q8 achieved 100%! #LLM#AI#MLhttps://github.com/chigkim/haystack-test
For inference, the best option right now is llama.cpp with quantized LLM in GGUF format. There are several high-lever wrappers around llama.cpp that makes it easy to use: ollama, vllama...
For inference with very big LLM and very small RAM, the only option is airLLM: it's slow, but you can run llama3-70b
For finetuning quantized LLM with LoRA, the only option afaik is also llama.cpp (look for "finetune"). It's a work in progress but usable and promising!
This kind of thing gets me going. These sort of reversals of agency. Also, the idea that things get caused by dreams. There is something in me that wants to explore all these things that don’t fit the not modernity
„Dzieje się tak dlatego, że LLM, niezależnie od tego, jak dobrze wyszkolony, nie potrafi ani abstrahować, ani rozumować jak człowiek. (...) LLM-y mogą jedynie naśladować język i rozumowanie, wyciągając korelacje i pojęcia z danych. Mogą często poprawnie naśladować ludzką komunikację, ale bez umiejętności internalizowania i z powodu olbrzymiego rozmiaru modelu nie ma gwarancji, że ich wybory będą bezpieczne albo etyczne" #ai#llm#wojsko
Here is a great summary or glossary doc about LLM by Aman Chadha. This long doc provides a summary of some of the main concepts related to LLM. This includes topics such as:
✅ Embeddings
✅ Vector database
✅ Prompt engineering
✅ Token
✅ RAG
✅ LLM performance evaluation
✅ Review main LLMs
Beyond the brain: Our intelligence leverages the power of culture and language. Channeling Ted Underwood and Francios Chollet, I argue that language models, despite their biases and lack of understanding –– are important tools for thinking. 🗣️🌍💡 cc @TedUnderwood https://leviathan.substack.com/p/beyond-the-brain
Ofcourse results needs to be verified and confirmed in practice but after reading the
MedGemini paper from Google there is no doubt in my mind AI will change the world of medicines. Not replacing people but augmenting them during diagnosis, operations and treatment of patients. https://arxiv.org/abs/2404.18416 #AI#medicines#generativeAI#LLM#GoogleGemini#MedGemini
If you are using LLMs through API tokens, or running locally, which UI do you use? I'm in the market for recommendations. Have tried llm and LibreChat but neither really stuck for me. #ai#llm
I really like the convention of using ✨ sparkle iconography as an “automagic” motif, e.g. to smart-adjust a photo or to automatically handle some setting. I hate that it has become the defacto iconography for generative AI. 🙁
Google provides a tool called PageSpeed Insights which gives a website some metrics to assess how well it is put together and how fast it loads. There are a lot of technical details but in general green scores are good, orange not great and red is bad.
I tried to ensure the tests were similar for each platform by choosing a page that shows a list of posts, like https://mastodon.social/explore.
The rest don’t seem to have prioritized performance or chose a software architecture that cannot be made to perform well on these metrics. It will be very interesting to see how that affects the cost of running large instances and the longevity of the platforms. Time will tell.
Je bosse au 4/5 sur les modèles de langage (LLM, parfois appelées IAs) et à 2/5 sur la robotique open hardware AMA (jlai.lu) French
Hello!...
Best printer 2024, best printer for home use, office use, printing labels, printer for school, homework printer you are a printer we are all printers (www.theverge.com)
After a full year of not thinking about printers, the best printer is still whatever random Brother laser printer that’s on sale.