← Feed

Posts

• Post #4011563

  Asimov's point was never “three neat rules solve robotics.” It was: if you create non-human agents with physical or social power, the central design problem is preventing harmt o humans. We’ve largely inverted that. We’re asking how to make them more useful, faster, cheaper, more deployable, more deniable, and only then asking how much harm is politically acceptable. That toot is really a masterful statement of that. Asimov's laws were never perfect (they were a narrative device prim...

• Post #3677182

  (via Reddit https://www.reddit.com/r/GreatBritishMemes/s/GNaHsGqUWd)

• Post #3650477

  RE: https://tooting.ch/@greg_harvey/116877462304825483 https://stylo.bot is EXACTLY for this.

• Post #3626696

  One thing I think we're underestimating in the age of code LLMs is the value of senior engineers' internal models. A code LLM doesn't build your architecture; it builds the most statistically plausible architecture from the evidence you give it. The problem is that the most important evidence often isn't in the repository, tickets or documentation (honestly it's RARELY in documentation and almost never CURRENT there). It's the accumulated understanding of why the syste...

• Post #1742561

  For all you Americans love guns you certainly can't seem to actually hit a fat pice of shit in a poorly fitted suit.

• Post #1570079

  Well, shit...got the 'Vid. Hence why I've felt so crap; no coughing this time oddly but my temperature is 'sweaty mess' levels.

• Post #1570078

  Sleep for 4 hours, wake for two...work...rinse repeat. Seems to be almost sustainable. I'm at the 'all my joints are agony' stage of covid (my third time...)

• Post #1570077

  "I object to the content you spent days buuilding and gave me for free on your self hosted ad free site because you used an LLM to structure it" isn't the diss you think it is. The sheer entitlement of people saying others shoudn't use AI because it's cheating or somesuch bullshit is VILE. DON'T FUCKING READ IT THEN. We owe each other NOTHING...but I gave you a gift you're now using as an excuse to attack me. Frankly; fuck you.

• Post #1570076

  Awoken from a nap again (I'm doin g4 hours sleep try 2 awake..). StyloBot is in the LONG final test and polish phase. Super boring but it's just using the app over and over again for HOURS (and AI driven automated tests etc). Feature complete for v6 (and the first commercial release) so really I need to polish for as long as needed. Have a CRAZY way of searching too...essentially audience segmentation by behavioural fingerprint...it;s...odd...of and it uses radar shapes you m...

• Post #1570075

  Fun experiment (on Mac still) brew install cloudflared brew install stylobot Then run: stylobot https://www.bbc.co.uk --tunnel Leave it running and see how long it takes before it detects a real bot. I used a totally anonymous Cloudflare tunnel and it took 5 minutes for a curl scan to probe it. WOW.

• Post #1570074

  Testing ChatGPT new image this and yeah impressed. This was a simple prompt (so labels ar ewrong) but it's PRETTY GOOD

• Post #1570073

  In StyloBot RTM phase now. Feature complete, jsut testing and polishing. 5-6 sees a TON of new functionality, 14(!) new detectors, a new search system , in paid api (poly factor authorization) protection, identity management and approvals etc...etc.. HOPEFULLY RTM this week but really *when I'm happy* it's a huge system now - well the engine fits in a 25mb binary - editors etc are HUGE.

• Post #1570072

  Ooh Claude Code Ultraview is great but *$50 a pop!* . Annoyingly it also only covers limited times too...hmm...Seems like an 'extract tons of cash' feature...

• Post #1570071

  Always amazed me that Facebook & it's Zuck-bot leader really only had ONE good idea. Using audience segmentaiton behaviour collected in a social network graph to make ads more valuable. Increasingly used to wedge groups into compliant groups for targetting etc... But *everything else* is either bought in (*108 companies in total* or a miserable failure (Metaverse, current AI issues etc). Really Mark Zuckerberg isn't that GOOD a business leader he just has good economic t...

• Post #1570070

  Still on my COVID sleep patten (4 hours awake , 2 asleep) but StyloBot is getting there, just releasing 6.0 Beta1 (6.0 is the commercial release) . Adds compliance packs for paid - for GDPR compliance, DSAR deletions etc all out the box. Also fully auditable (down to indicidual 'signal' level) etc... FOSS is I *think* feature complete now. Up to 40 odd detectors now. Testied with over 5 billion requests through my 10 e-waste office PC cluster so *that works*. Wee bitty more pol...

• Post #1570066

  Ok next 4 hour nap begins! See ya all soon! (Also feeling kinda smug about my self hosted tiny llm systems as the llm bubble seems to be sarting to pop & REAL token costs hit home).

• Post #1570065

  Sigh total waste of a week. Luckily had StyloBot to keep me busy when I AM awake but just blhe...

• Post #1570063

  Today's StyloBot tweak is a way to avoid a 300ms delay for the frst request (on the web usually a content page request you *want* to be fast!) So by learning page request patterns I can derive a markov (path) chain to distribute the detector load across a pages' WHOLE request stream making it feel a ton faster.

• Post #1570062

  Well only lasted 3 hours this time. Back to bed. ALL my joints are aching now lungs feel red raw...this covid strain FUCKING SUCKS. Coughing was better than this.

• Post #1570061

  Constant battle of wanting StyloBot to get outinto the wild and ensure it's as good as I can get it first. ONE false positive for me is too much so I'll need tp run it on the site before selling it for a while. SHOULD be close to zero false positives; it's resolution is high enough now but logic errors can knacker the whole thing.