janekm

Well, how about this one... Opus 5.5 wrote a raytracer, IK animation, and pelican riding a bicycle for a novel VM with new instruction set and micro-GPU instruction set (DB32), in 4kb! https://demobench.janekm.com/studio?demo=pelican

I often see charitable groups from the Sikh community in central London serving the homeless. Interesting to hear the background around that! https://asiasamachar.com/2023/10/10/meet-the-sikh-community-...

Well the code is there: https://github.com/voideditor/void/tree/main/extensions/void But it's super basic right now... Like literally just the bare minimum to make an LLM call in a side window and a view to apply/reject diffs. Pretty much a weekend project at this point.

Regarding hosting, it's a big pain point for companies building these systems out of China as the "great firewall" makes it really hard to have reliable connectivity between the Chinese Internet (and datacenters) and outside. The ap-east-1 zone of AWS, based in Hong Kong, is very popular in that context as it tends to be accessible from both sides (so cost is fairly typical of AWS costs in that case). Alternatively Alibaba also have cloud datacenters in Hong Kong with similar properties.

Something worth keeping in mind is that Chinese factories that makes things like IP cameras don't typically have their own (significant) R&D departments and instead work from examples provided by the makers of the chipsets. Sometimes there will be another company involved that provide an "IoT ecosystem" (Tuya is a major player here: https://www.tuya.com/solution/hardware/ip-camera ). So it's not necessarily surprising that the backend still works even if the manufacturer (Amazon brands are not typically the factory themselves, for practical reasons frequently changing "brand names" are used) is no longer present.

A club is one of the few places where this strategy won't work, unless there's an emergency (which the bouncers would obviously know about) the only people who would show up are from the sound and light company whom the bouncers already know.

It's the curse of being a company with a $xxxB main product line. It becomes really hard to justify projects that could at best contribute a few million to your bottom line...

You might be interested in knowing that LCSC (same company as JLCPCB) are about to offer proper panel / membrane keypad prototyping: https://www.lcsc.com/panel-printing

That's going to be something similar to IPAdapter FaceID: https://ipadapterfaceid.com Basically you use a facial structure representation that you'd use for face recognition (which of course Apple already compute on all your photos) together with some additional feature representations to guide the image generation. No need for additional fine-tuning. A similar approach could likely be used for handwriting generation.

It's almost as if politicians are less motivated to enact legislation that is inconvenient to politicians.

Pick the ones that give a delivery deadline (5/7 days). No problems at all (at least from UK), always arrives ahead of schedule.

Came to say this too... I was really pleasantly surprised by how easy it was to get a full dev & build & test environment set up with Visual Studio Code, and how well their little WCH-LinkE debugger works (both SWD and UART in one, Segger could learn from that...). Now I'm experimenting with their BLE variant, which is a lot more complex, but still simpler than what I'm used to.

Well, here's "Not James Hoffman" having a go: https://youtu.be/-yB6Fb05QRo?si=PqhwiRsRnz39LOYJ

The Kino introductory blog post makes quite clear that all they do is apply one of a set of LUTs that ship with their app. Personally I'd be interested in one that tried to apply colouring more "intelligently", e.g. detecting faces etc and applying appropriate settings.

It's tough in this case, because if you incentivise just to save cost, it could always route you to the cheapest LLM but the quality would suffer...

Yes, but this scenario is a bit different. If you ask a connector factory to custom-make you a connector it's common that they are allowed to add it to their catalogue, though you can of course stipulate something else in the contract. You'll see this in the electronics markets in Shenzhen for example... they still have a lot of the connectors and switches from MP3 players, which I believe was one of the early boom industries in the region.

There's over 182 domestic routes flown in the UK (Heathrow - Edinburgh is the most popular one): https://commonslibrary.parliament.uk/domestic-flights-in-the...

But you can only really make those profitably for a few industries (military, medical, seismic come to mind). The EU does have the chip fabs for those industries, of course...

60g/l acetic acid ... that's literally just slightly strong household vinegar. (safety precautions, especially not breathing it in, are definitely advisable though, it's a fairly strong acid)

It seems we'll have convincing AI video impersonation before practical quantum cryptanalysis, unfortunately...

It's more that the LLM non-local hosting (and training) community (on A100/H100) is big enough that it's more profitable for Nvidia to not release such prosumer GPUs ;)

None of the open source models I've seen are as "well-rounded", production ready as Eleven Labs. Though for example bark is really great at prosody: https://suno-ai.notion.site/Bark-Examples-5edae8b02a604b54a4... And piper isn't bad at speech quality: https://rhasspy.github.io/piper-samples/

We might only be a few papers away from a good open source Elevenlabs competitor.

Yeah, sadly it'd cost about $100 to get a book per month... Not quite competitive with Audible yet, but give it a year perhaps, or a few iterations of the open source models... (fixed the link)

thanks!

Elevenlabs is a lot closer to compelling audiobook narration (needs a better way to deal with multiple characters in a story without manual use of multiple voices): https://pub-a24da573c61f4b2d905bdebb2d0ecf88.r2.dev/ElevenLa... (an H.G.Wells example I just asked it to read).

I think the idea is "figure out a list of people who appear to have shared interests whom you might enjoy talking to", but the chatgpt reference does appear odd indeed

It sounds fascinating! Is there anything one could read to figure out more about how this is being done (From reading the docs by the "Teleprompter"s right)?

I took that comment to refer to the scrappy original AirBnB, not the one we see today (I can't recall exactly how the first implementation was done but I do recall that it was something really simple along the lines of an excel spreadsheet and email). It's in that space of proof of concept implementations that the LLMs can really accelerate (and, potentially, as a "pair programmer" kind of role).

I for one am willing to offer $1 to buy a 0.0000001% stake in GPT-Unicorn. Anyone else? :)

Giving moderators the power to arbitrarily ban users they deem to have produced content by AI is not likely to work for long though. Perhaps today it's still feasible to note the "vibe" for GPT, but that's not likely to be possible for long (it's probably an artefact of the RLHF process anyway). Then what? You'll end up with essentially arbitrary application of the ban hammer based on the moderators' "feelings".

I certainly don't have the answers (and let's face it it's likely going to be a problem here too).