According to the writeup; there are two methods: it is possible for an extension to mark some parts of itself as 'web accessible'; and linkedin has assembled at least one characteristic file for 6,1000-odd extension IDs and attempts to fetch it to confirm/deny the extension's presence.

The other is based on the fact that the whole point of many extensions is to modify the site in some way; but the site normally has largely unfettered access to inspect itself, so they have theirs set up to walk the entire DOM looking for any references to "chrome-extension://" and snagging the IDs if found.

Not exactly a 'declare installed extensions'; but it looks like, out of some combination of supporting the use cases where an extension and page actively interact by design and either not wanting the possibility or not wanting the complexity of trying to enable 'invisible' edits(presumably some sort of 'shadow' DOM mechanism where as far as the site and everything delivered with it knows only its unedited DOM and resources exist; but the one the user sees is an extension-modified copy of that one, which sounds like it could get messy), inferential attacks are fairly easy and powerful.

Where exactly does supporting 3 people on $133k/year count as 'upper middle class'? You could be doing a lot worse, and many are; but that's not just tons of money in a HCOL area; and that's also lower than twice the median salary for full time employees with bachelor's degrees; so you are calling either a single income household doing a bit better than median or a dual income one doing worse 'upper middle class'; which seems pretty ambitious.

It's hard to train a cow to keep its cellphone charged; so you can see why you need something to keep their tracking collars topped up.

Realistically, the status quo has arguably outrun the dystopia there. Your phone already does far more than anything you could get into the power envelope of a bracelet or embedded chip implant, and if for some reason you've raised enough eyebrows that you'd be hauled in for an RFID read DNA is a pretty indelible identifier.

It's not 100% ironclad; but penetration is broad enough that you've basically got the majority carrying highly fingerprintable RF beacons and the minority standing out for their relative radio silence and attempts to deal in cash. Expensive and uncomfortable ankle trackers are good business and feel nice and punitive, just to remind the wrong sort of people we aren't happy with them; but you don't really need to impose a surveillance society when it will build itself for you.

My (admittedly anecdotal from the totally unscientific sample of random stuff I've had reason to work on) impression is that some 'shared' BMC ports had oddities related to network controller sideband interface speeds, since NC-SI is what the BMC is depending on if the NIC is on someone else's PCIe root. It's not like the BMC actually needs a faster link for much(normal management traffic probably doesn't fill 10mb and mounting virtual media may be literally once-in-a-lifetime) so the actual speed of the NC-SI interface was not a burning priority; but it left things up to the NIC whether it would support remaining at gigabit speeds and just quietly slipping the trickle of shared traffic in(presumably slightly more complex; but seems to be what the newer ones do) or if it would knock the link rate down visibly to simplify the case.

You see little echoes of similar behavior elsewhere. The intel desktop and laptop NICs that support 'vPRO' will be GB or 2.5GB when the system is on; but quietly drop back to 10 or 10/100 when it is off and it's just the management engine listening. Some enterprise vendor USB docks do similar things; looks like a normal USB NIC when the OS is up; but drops to a lower speed and operates quietly over, I think, some sort of oddball vendor-defined messages if one of their systems is plugged in but off.

I assume that, as an exercise, getting 5 simultaneous introductions working makes for a better paper; but is there a reason why you would want that in practice? Especially if there is any wobble in the ratios either randomly, across generations, or in the presence of certain environmental conditions that tweak the plant's metabolism one way or another that sounds like it would be a real pain in the ass to have to re-balance (and, if different patients are deemed to need different combinations even a perfectly stable plant is going to need re-balancing of the outputs) vs. very specifically going for a specific target output per-plant(or e. coli or yeast or whatever is easiest to bioreactor) and then just mixing to taste after purification. Is there some advantage I'm not seeing?

I realize that there are cases where some plant-sourced pharmacological effect looks like it is actually driven not by the identified 'active ingredient'; but by dozens or hundreds of assorted things, and in that case you just have to live with the complexity if you get better results with that than with purified isolates; but if you are deliberately engineering for very specific outputs why a mix of 5?

output is about 6 tokens/s with 16k context window i'm not having any issues since it went live this afternoon. it's not sparkling like opus 4.5/6 but gets the job done

i generally send it a voice note via telegram while driving and then check back in like 1-2 min, or it is sending me a reminder about something on our shared calendar. it's still faster than texting my buddy about making plans for this weekend or whatever.

I'm using a $200 used ~5 year old (from the ebay listing) HP EliteDesk 805 G6 DM Desktop Ryzen 5 PRO 4650GE 3.3GHz 32GB RAM 512GB SSD WiFi in cpu mode... you don't need a gpu to run single user local LLM... just a bunch of ram. This isn't 2022 anymore

It's about 5 tokens/second which is totally fine for an async assistant. 20 tokens/second is about the lower limit for usable in realtime. You can also set it up to use a smaller model for quick questions (what are the next 6 items on my calendar/to-do list?) and drop through to the bigger slower model for harder questions (can you add this feature to my internal ticketing system and redeploy?)

I ordered 64gb of ram about an hour ago and i'm planning on running either qwen 35B-A3B 8 bit or 122B-A10B 3 bit in fully offline mode.
 
  >the actual cost of 'running the AI.'
 
is a fixed $200 cost (ram upgrade) + electricity

I cancelled my subscription overnight, and I'm using the free credits they gave me to wrap up some things and transition away. I am not going to be locked into someone's walled garden again.

Immunology, presumably.

The only donor bodies that aren't going to treat the transplant as an act of war are clones or heavily immunosuppressed; and it's probably more plausible to assume that you'll be able to clone a human like a sheep than assume that you'll be making some fundamental breakthroughs in immunology to deal more elegantly with unmatched hosts.

I can see the utility of having spare organs in certain emergencies; but how much life extension would you actually get even if the sort of neurosurgery involved in removing a brain and reattaching it to a new host's spinal cord were viable? Is the theory that the assorted ghastly flavors of neurodegeneration are actually to be blamed on older organs and everything will be fine; or is this just a very expensive way to ensure that you skip the various ways peripheral organs can kill you and are assured to be the spryest patient in the dementia ward?

Even if I could in theory change my email address, there is so much depending on it, it would be huge trouble to change it. For example, if you are in the UK with settled status, you need to make sure you keep your email address and phone number. Result: Iâ(TM)m paying BT £7.50 every month for three email addresses (up to 10 actually, but I only need three).