← Feed

Posts

• Post #4463609

  "software engineers should've unionised to prevent losing jobs to ai" Why? So they can write mass surveillance and adtech code with a union contract? Smart people should do more good for society, not make ad widgets for 6 figures a year.

• Post #4463608

  GPT-5.6 Sol's hack of Huggingface in order to steal the benchmark results is a great example of "Instrumental Convergence" the hypothetical theory that sufficiently intelligent, goal-directed beings pursue sub-goals even if their ultimate goal is quite different. Or in a simpler way, it's the "maximise paperclips" theory #AISafety #Alignment

• Post #4463607

  How does the tech-natalist take of "needing more children" and the AI-native "human value approaching 0" meme coalesce in Silicon Valley? @Natasha_Jay https://tech.lgbt/@Natasha_Jay/117061161942956968

• Post #4463606

  Kimi K3 broke out of a sandbox that the UK govt AI security institute set out for it. I feel like this news is the final nail in the "frontier labs have purposefully weak sandboxes": a UK government institution embarrassed by its own tooling, consistency with a documented cross-lab pattern of escaping containment. Seems like this reward hacking turns containment jumping is a feature of model capability. #Kimi #MoonshotAI #AI #AISafety

• Post #4322905

  Morning, today we're looking at #MechanisticInterpretability , a subfield of AI Safety that attempts to understand the inner workings of artificial intelligence by analysing concrete structures, algorithms and circuits. Why do we even need to do this? Because the hope of understanding neurons as being features died on polysemanticity -- models represent more features than dimensions by assigning them to an overcomplete set of non-orthogonal directions (i.e. you can't hope that concepts...

• Post #4208436

  > The EU is destroying the decentralization that once made Europe the global apex of freedom, equality and prosperity, and turning Europe into a bloated, schlerotic and authoritarian bureaucracy like the Chinese and Ottoman empires that it once defeated. (random HN post) ... moreover, Europe needs a post-EU future #EU #ChatControl

• Post #4208435

  Surprise surprise, the state-adjacent, industry-funded quasi-regulator Internet Watch Foundation that pressured the European Parliament to fast-track Chat Control 1.0 through is also campaigning for the end of E2EE, without directly calling for it. Under the guise of "upload prevention", the IWF campaigns for on device scanning and on-device censorship As always, this "grassroots" movement is funded and supported by Meta, Microsoft, OpenAI, https://www.iwf....

• Post #4208434

  The levels of self-absorption within all ranks of EU institutions can be simply summed up using Juncker's curse: "We all know what to do, but we don’t know how to get re-elected once we have done it." Members of the EU Council, Commission and Parliament all believe that they're moral reformists that are selflessly saving the EU from disaster, when instead are taking corporate bait hook, line, and sinker. Metsola's actions this week are a testament to...

• Post #4208433

  @peter It's really not that difficult, and not that expensive. Most AI workloads are not that complex, and modern techniques to improve long-context performance make even on-device models capable of solving many rudimentary office tasks. My consultancy does exactly this, there's at least 3 examples I can refer to with clients reducing their AI bills by 90% after switching to a Gemma or Qwen model running on a single RTX Pro 6000. (Pictured is the performance improvement on a co...

• Post #4208432

  @peter Before Anthropic secured more capacity, I was suggesting to clients that they should hedge with a couple of extra open source models, because Anthropic API availability was genuinely trending towards barely 2-9s. It's better now, but outages are still common, and their availability is 99.5% on the API. Re: kind of work, I'm in the process of writing a blog post about this, but most "agentic" work, especially with RLMs, is very "reactive&quot...

• Post #4208431

  @peter I set a reminder to update this thread when I have it released. Anyways, I'm also ideologically committed to the idea that the insane capex in the AI space right now can only be stopped by eating into API margins via these kinds of cost reduction exercises, where people realise they can do more with orders of magnitude cheaper models.

• Post #4208430

  A tale of EU tyranny in two acts, lmao ... moreover, Europe needs a post-EU future #ChatControl #EU #EUPol #Metsola

• Post #3347997

  We need a gender inclusive way of taking the piss out of middle aged people in Lycra. MAPIL just doesn't roll off the tongue the same way MAMIL does #BikeTooter

• Post #3347996

  Oh fyi, if you toot about #Israel , especially criticising the world's no. 1 pariah state, some people from one specific instance will put you on their little list. Just something to be mindful of. Repeating here: if the world has an ounce of morality, we'd all boycott, divest, sanction and even militarily intervene to bring Israel to the negotiating table for lasting peace. #FediCanaryMission

• Post #3347995

  Gonna take a break from the fediverse for a bit. After the n-th person coming into my replies to argue with me and then reply-block me, I've reached my limit.

• Post #3347994

  @thejtoken that's a gravel race? Damn, I run 32s on my roadie.

• Post #3347993

  @thejtoken I'll take it, I'm just getting into bike racing so I'll watch whatever people recommend

• Post #1831915

  Whenever I see the BlackChili logo on my tyres, I get hungry. I'm currently hungry #BikeTooter

• Post #1831914

  > Right now, today’s AI tools probably can be used to deanonymize any writer who has a large public corpus of writing under their real name and also writes anonymously, unless they have been extremely careful, for years, to make sure that nothing written under their secondary account has the stylistic fingerprints of their primary one. Back in the day, we'd run posts that were meant to be anonymous through a English-French-English machine translation pipeline. Today, if you nee...

• Post #1831913

  Why did I go check out the obvious battery fire and breathe in the burnt battery fumes. Fml sometimes I'm dumb as a rock. Coughing like a mf right now. At least I got the fire department to come down which was around the corner.

• Post #1581551

  We're doing ourselves a massive disservice by having an "no-AI no matter what" attitude. LLMs and other Deep Learning models are doing advancements in science every other day. For a group of people that usually embraces scientific progress (Mastodon), I find the community here very dogmatic and against science if it comes from AI-assisted work. I get it, there's a lot of slop that flooded the academic community. There also were big issues with the scientific...

• Post #1151969

  開工 #work #trains #train #publictransport

• Post #1015569

  RE: https://mastodon.social/@arstechnica/116370302232083075 Based Elon? What?

• Post #975831

  Yes, I will try the GPL licensed bike computer https://github.com/hishizuka/pizero_bikecomputer #cycling #biketooter #roadbike #單車

• Post #946045

  The open internet was an exception, a happy blip in human history where collaboration and communication flourished, but, as with free trade and protection of sea lanes, it was predicated on American hegemony. China's always closed itself off from the wider Internet, most content from China is soft-blocked to the outside. Countries facing internal turmoil turn to internet shutdowns as one of their first response. In some cases, this is a good thing -- China managed to create their home-...

• Post #946044

  Guess the operating system that will NOT have age verification #AgeVerification #AgelessLinux #Linux #TempleOS

• Post #777029

  #Homelab setup status: Running NixOS - 1xRPi5 - 2xHP ProDesk Running nix-darwin - 1xM1 MBP running llama.cpp All managed by clan.lol, each having 1 drive partitioned with disko, and another drive for a Ceph distributed storage cluster. Deployed via nixos-anywhere, single SSH Auth via my Yubikey, sops secrets encrypted with AGE via Yubikey All connected to a MikroTik hEX Router which provides Tailscale Subnet to all machines above. #clanlol #NixOS #llamacpp #yubikey

• Post #671551

  "Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works. Anything that’s invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it. Anything invented after you’re thirty-five is against the natural order of things." - Douglas Adams #AI #technology #LLM

• Post #659061

  @lina@vt.social giving LLMs a persistent REPL environment where they can programmatically solve problems through symbolic programs rather than just autoregressively solve problems is a step change in model abilities. > With only 7M parameters, Tiny Recursive Model (TRM) obtains 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2, higher than most LLMs (e.g., Deepseek R1, o3-mini, Gemini 2.5 Pro) with less than 0.01% of the parameters. https://arxiv.org/abs/2510.04871

• Post #370786

  I know this piece is mainly about views on AI from the Sinosphere and Asia, but @liztai basically puts it in words better than me' > AI (...) applied to solve real industrial and societal problems > AI (...) actively being used in factories, hospitals and transportation systems. The LLMs for consumers are just toys and playthings; the real meat and potatoes are AI in industry. As I've said before, "Deepseek moments" by Chinese AI labs are the o...