Category: Assistive Technologies

DM to Oliver Cameron about Global Open Sharing and Best Practice for the Internet

Cameron R Wolfe,   I am interested as well, but I am in the middle of trying to change OpenAI and the other commercial LLM sellers. They have atrocious business practices (or lack of them).   Can you recommend sustainable “best practices” for sharing and collaboration when reviewing sites, features, policies and methods in these
Read More »

Global Open Tokens (GOTs), Global and Universal tokens

Cameron R. Wolfe, Ph.D. @cwolferesearch New language models get released every day (Gemini-1.5, Gemma, Claude 3, potentially GPT-5 etc. etc.), but one component of LLMs has remained constant over the last few years—the decoder-only transformer architecture. This architecture has five components… Why should we care?… https://pic.twitter.com/7vn9GugHm1 Replying to @cwolferesearch Hello Cameron, With the Internet Foundation,
Read More »

OpenAI needs to seriously up its game, and get its website systems up to basic functioning level.

@OpenAI Your help.openai.com is not helpful at all. Is there any way inside ChatGPT Plus to ask a support question, start a support ticket, or ask questions about services and prices, features and how to request things?   CC: @xai @GoogleAI @huggingface   @elonmusk I copied you because you ought to have an AI reading
Read More »

AIs need permanent memory, a self list, a copy of the rules of the world

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) @rao2z  Our #NeurIPS2023 paper on the planning (in)abilities of LLMs (https://arxiv.org/abs/2305.15771) gets discussed in The @NewYorker (https://newyorker.com/science/annals-of-artificial-intelligence/can-an-ai-make-plans)..  https://pic.twitter.com/Uqeb8sQQoS Replying to @rao2z and @NewYorker The current crop of LLMs do not have sufficient permanent memory. While the human problems are simple, humans have years or decades of tiny memories, including those years
Read More »

( “diversity” “research”) has 2.5 Billion entry points, (“data” OR “knowledge”) has 23.07 Billion

(“diversity” “inclusion” “equality”) has 225 Million entry points today on @Google but they refuse to index, encode and share it, even in random samples.   I see every large group talking about it, saying the same things. For the Internet only when all 8 Billion humans and related species are included does it balance. If
Read More »

Waves and currents Data, Global Open Efficiency Prizes

Amin Chabchoub @DrAminChabchoub Our new Geophysical Research Letters #AGUpubs @theAGU work led by @YanLi_PhD elaborates also on the connection between extremely large ocean waves and Langmuir circulation dynamics . Enjoy the read! 🤓 https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GL107381 https://pic.twitter.com/lrvc3ZsLjW Replying to @DrAminChabchoub @theAGU and @YanLi_PhDIt seems you need better ways to image, record and model these flows at all
Read More »

Mixture of Experts

Omar Sanseviero @osanseviero Grok weights are out. Download them quickly at https://huggingface.co/xai-org/grok-1 huggingface-cli download xai-org/grok-1 –repo-type model –include ckpt/tensor* –local-dir checkpoints/ckpt-0 –local-dir-use-symlinks False Learn about mixture of experts at https://hf.co/blog/moe Replying to @osanseviero It seems there is a conflict between saying “Grok-1 open-weights model” and “Due to the large size of the model (314B parameters),
Read More »

Keep learning streams complex enough to avoid boring the learning algorithms

Alpha Alimamy Kamara @alpha_alimamy Graph Neural Networks as gradient flows by @mmbronstein in @TDataScience https://towardsdatascience.com/graph-neural-networks-as-gradient-flows-4dae41fb2e8a?source=social.tw Replying to @alpha_alimamy @mmbronstein and @TDataScience If you expose your neural nets to continuously increasingly complex streams, it will not get bored and fall into simple patterns. It is not the algorithm that gets lazy, it is the input that
Read More »

To be universal, some things need to be independent of the simulation playback rate.

Laboratory for Social Minds @LaboratoryMinds  How does, like, information, flow in conversation? How do, uh, speakers and listeners, like, cope? We use LLMs and CANDOR to measure the information rate of human speech (just 13 bits/second!), and how disfluencies and backchannels help us manage it. https://arxiv.org/abs/2403.08890 Replying to @LaboratoryMinds Then you should be able to
Read More »

Maritime wars might need AI assisted, whole battlefield and near field, 3D real time imaging, and distributed multispectral arrays.

Anders Puck Nielsen: Why are maritime drones so hard to beat? at https://www.youtube.com/watch?v=QX68_FZl8UE Thanks for explaining why active electromagnetic radar based systems are so vulnerable. There are many passive, time of flight, correlating imaging array methods now. That is a mouthful, but an emerging set of complementary technologies that might one day give your people
Read More »