Reposted by GusJoshua White @jrw14.whnc.me · 04/10/2026I do not understand the people who willingly use Meta's products. www.wired.com/story/muse-c...wired.comMuse Creates Detailed Profiles of All Your Friends and FamilyMillions have downloaded Meta’s AI agent Muse. But getting it to do your bidding comes with privacy costs. 1225
Gus @random-gus.bsky.social · 02/10/2026Hedging sounds to me like a natural consequence of a LLM cosplaying Jev. I really want more architectural details from Typesafe, it seems classification is way more integrated during the entire training process than the clones are able to do now 000
Gus @random-gus.bsky.social · 02/10/2026Not sure it’s new, but using codex to compile Latex I saw it writing Lua scripts which I found really interesting 001
Reposted by Guspmamtraveller.bsky.social @pmamtraveller.bsky.social · 30/09/2026By Will McPhail 565007918
Gus @random-gus.bsky.social · 30/09/2026@jevbot.bsky.social 01101000 01101111 01110111 00100000 01100001 01110010 01100101 00100000 01111001 01101111 01110101 00111111 110
Gus @random-gus.bsky.social · 30/09/2026Dots prompts codex instead of you?? The prompting on your behalf is free, whatever codex does is not??? 012
Gus @random-gus.bsky.social · 28/09/2026The “””most influential””” man in AI…. Spells out dot com and claims he can’t refuse the money because there is no UI action he can do? 021
Gus @random-gus.bsky.social · 25/09/2026Don’t kill humans is what a lot of people think when talking about alignment. P(doom)!?? Yes it is too far and a mess of a topic to discuss. But how should labs measure how aligned to my objects the models are? To yours? They are not the same. 100
Gus @random-gus.bsky.social · 25/09/20262 different issues I think, the term itself and the idea. I honestly think we need a new term alignment means nothing anymore or the least common denominator is: please dont wipe out humanity. As how to measure and guarantee this individual/org goal I find it much harder, because it is unique? 100
Gus @random-gus.bsky.social · 25/09/2026Yeah there is no single form of alignment, an issue I always had with the term. But due to all the baggage maybe we should adopt a new term for personal/organizational “straightening” 120
Reposted by GusCarbonadoks @carbonadoks.com · 24/09/2026Pleometrics PC-98 P(Doom) video is so good. 811822
Gus @random-gus.bsky.social · 23/09/2026As a user who paid for one year upfront I feel the same and regret paying. I used to think they were serious and cared about safety. 010
Gus @random-gus.bsky.social · 21/09/2026Don’t build a massive user profile based on everything stored on my computer and turn that into ad targeting not an option? 120
Gus @random-gus.bsky.social · 21/09/2026If Jev is a pre-trained classifier when will we see Kay the pre-trained regressor? 000
Gus @random-gus.bsky.social · 21/09/2026Generative pre-trained transformer was a bit of a revolution. Now we are seeing a pre-trained classifier, Jev, and people are really excited. I like that people are using and learning about “traditional” ML. Go learn scikit-learn, is much easier than PyTorch and runs on your laptop, no cuda needed 110
Gus @random-gus.bsky.social · 19/09/2026North America includes Mexico as countless treaties, accords and organizations that are composed of these 3 countries 000
Gus @random-gus.bsky.social · 19/09/2026That was the idea. Didn’t say it was fair 😁 But can Jev plan ahead or will it trap itself? Possibly the 2d version would be harder for Jev, easier to find a way out in 3d if you are a bot. 010
Gus @random-gus.bsky.social · 17/09/2026True, but I like and will steal your bigger metaphor of camera and smartphone but the results depends on who is using either 010
Gus @random-gus.bsky.social · 17/09/2026An amateur spending thousands on camera gear would still take meh pictures. An amateur with Astra/Fable and the coolest harness will make meh software?? 100
Gus @random-gus.bsky.social · 17/09/2026Yeah that if is a gigantic if. How many times they have baited and switched before. We know what Meta is and what they do. Why would they do anything different in the long run? Zero chance. Frog, water, boil etc 000
Gus @random-gus.bsky.social · 17/09/2026I have to finish writing an academic paper this month. So this is perfect rabbit hole that I should fall into now. But I will later 110
Gus @random-gus.bsky.social · 17/09/2026Love the idea. Last week I was wondering if I should ask astra to rebuild the old ios game flight control for me. You did it in 3d 120
Gus @random-gus.bsky.social · 17/09/2026Now I’m intrigued with what comes next. Because parallel computing is not limited to classifiers. Next a regression model and finally a generators??? 000
Gus @random-gus.bsky.social · 17/09/2026Got it now, haven’t used yet. Now I get it Jev is a classifier, apparently an amazing one out of the box with no fine tuning. That’s why the output is free. Is a class and confidence score 120
Gus @random-gus.bsky.social · 17/09/2026I don’t have access yet. But found their description a bit limiting “Jev is designed for common-sense judgments: classifying content, routing requests, scoring responses, and evaluating information. You can combine these decisions within your code to form larger workflows.” I don’t really need that 200
Gus @random-gus.bsky.social · 16/09/2026It’s crazy that one graph ruined the entire category Like: I once say a pizza graph SOOO bad that now I don’t even eat pizza anymore 000
Gus @random-gus.bsky.social · 14/09/2026using AI to code is the same as disagreeing with Field medalist about how huge AI labs are using millennium problems as marketing? Tao wrote about how he uses AI not long ago. Their latest post wasn’t against AI as a whole, just that solving those problems is not the same as advancing the field 100
Gus @random-gus.bsky.social · 14/09/2026The article says what it is. Stock based compensation, a lot of tech companies do the same kind of adjustment 000