This is fascinating work by Anthropic to try to engineer safety and mistakes from inside the models.
A research paper by Anthropic details how decomposing groups of neurons in a #neural network into interpretable “features” may improve safety by enabling monitoring of #LLMs.
The hubs and I were having a discussion about Apple continually releasing new iPhones without any actual innovation. That lead us to a few questions, that I, of course, said, Well, I'll ask Mastodon about it! So here goes:
Have we reached as far with Phone Technology, as it exists as a rectangle piece of hardware in your pocket, as we can?
Is the future in software?
What about a better version of Google Glass?
Are we heading towards implants?
Please don't mention anything with a see through screen, literally no one wants other people to see what they are doing on their phone! LOL
@RickiTarr 2/ So everything gets much smaller. Implants and #neural interfaces are one possible future, but there are many unknowns along that path. Instead, think #wearables. An earring maybe.
(2) #Software (more generally, re-programability) offers plenty of opportunity for #innovation. #Middleware will be key. For example, self-organizing distributed virtual #computing systems will be far more powerful than any standalone device.
Defining #neural#modularity is hard: much history. We used toy ANNs to show structural and functional definitions not tightly related, resource constraints important, and we need to start thinking about temporal dynamics.
this may be a silly little thing, but it shows how using a social media site for 16 years affects muscle memory: i still hit the star instead of the ribbon to bookmark a toot.
back on #twitter one of my followers said they have a hard time relating to the ui; while another said they had suburban (and by that i mean white) gated community anxiety when looking at the fediverse.
and this is why the corporatocracy went all in on social media: they know they can affect us as a #neural level
Our innovation start up
https://ibb.co/ScMkgn1...
Swiss-French team gets paraplegic walking again with thought-controlled implants (www.rfi.fr)
In a world first, electronic implants developed by Swiss and French experts have allowed a paralysed man to walk again simply by thinking about it.
Neural Networks Need Data to Learn. Even If It’s Fake. (www.quantamagazine.org)
I’ve been following the development of the next Stable Diffusion model, and I’ve seen this approach mentioned....