happycube

(side-note: You can get Sandisk stuff directly from Western Digital.)

The bug report shows a Mali GPU, which rules out Snapdragon (Adreno GPU). Also, the phone model when googled shows the S20 FE, which has Exynos.

N4020 is quite old too. The N95/100/200, single channel memory aside, holds up pretty well against even 6th gen i5's and has AVX2.

Some of it's being fed by the companies themselves going "oooh, look how dangerous our new product is! But don't worry, we'll keep it safe..."

Also note that for the last couple of years the Pi2 is really the 3-- (a 3B without wireless) - for logistics reasons they upgraded the SOC.

The problem is 'weird stuff' on chai included suicide, which sadly showed it needed better/more guardrails.

How far we've come from GPT2...

I got the impression that they only have a USB2 uplink, which is unfortunate for a $380 device in 2023.

A 24-core Ryzen with an 8C zen4+3d sram and a 16C zen4c chip might be quite interesting. There are probably already a few in AMD's labs ;)

50% of the wings on a 787 are carbon fiber. Which was tested far above loads before failing.

When corners aren't cut building the planes, and Boeing has never cut corn... oh, crud.

I think there are searchable .pdf's on archive, and OCR dumps (which aren't completely accurate but butter than nuthting)

It already is, and everything through March 2023 was archived by pushshift and there are torrents floating about.

It's about 2TB in zstd compression, so finding the needles will be interesting - but it'll probably be much easier years from now.

This sounds eerily like LLM hallucination:

--- In one of Gazzaniga’s experiments, researchers presented the word “walk” to a patient’s right brain only. The patient immediately responded to the request and stood up and started to leave the van in which the testing was taking place. When the patient’s left brain, which is responsible for language, was asked why he got up to walk, the interpreter came up with a plausible but completely incorrect explanation: “I’m going into the house to get a Coke.”

In another exercise, the word “laugh” was presented to the right brain and the patient complied. When asked why she was laughing, her left brain responded by cracking a joke: “You guys come up and test us each month. What a way to make a living!” Remember, the correct answer here would have been, “I got up because you asked me to,” and “I laughed because you asked me to,” but since the left brain didn’t have access to these requests, it made up an answer and believed it rather than saying, “I don’t know why I just did that.” ---

And even if they did need it, they won't need it nearly as much the way things are going now...

Yup. It's awesome and rubbish at the same time.

Aside from interfacing with ISA cards emulation is probably better - an old mid/high end phone with a portable bluetooth keyboard and limbo x86 (via f-droid) sounds promising.

For laptops, there are some lovely cheap little Chromebooks out there that are near/past their update expiration date and if you want a slight challenge you can get an ARM one like the Asus C101P.

But there's something about having real hardware to do things like this...

I haven't done a lot per se, but 480Mb USB would be nice.

There are two major advantages though - you can retrain them, and they're not on the guardrails that the commercial models have.

Yeah - so much of writer's block is looking at an empty page/screen. It's very often easier to fix something bad than to write something good.

sigh yeah - GM's replacing the Bolt with the 2-ft larger Equinox for instance.

I'd rather have a revised Spark EV with rear wheel drive and 200mi range myself.

I forget if Langchain can do that, but something along those lines will exist if it doesn't already, it's too obvious not to, and too important for the free ChatGPT equivalents which will be popping up over a matter of days/weeks now that truly free versions of llama are coming out.

TL;DR the puck should get there very soon, not just Really Soon Now.

(I think there was one final intel round that had a reasonable keyboard, but I don't keep track enough to know if it was as good in other ways...)

And far more expensive than ChatGPT via API, so it makes sense to use ChatGPT3.5, or the locally run equivalents once they get as good, as much as possible.

De facto, yes. There was no way the weights wouldn't be posted everywhere once they went out to that many people.

It's a poor use of AI really that doesn't add anything to the product, and was instead clearly there to add something to the stock price to cash in on the hype.

If it was actually opt-in and helped people make better snaps...

(I'm well outside their key demo now, but their push messages seem out of touch even taking that into account. No, I'm not interested in lenses or adding people I've never had any reason to hear of...)

edit: Thinking about it a bit more, after their initial thing they've generally been missing boats... they could have made a nice TikTok clone with the snap map as a bonus - but most of the vaguely interesting snaps on the map don't have user info, and subscribing to the interesting users seems bugged on top of that.

Considering Tom Clancy became Tom Clancy(tm) several years before he died, heavily using ghostwriters...

There'll be an AI Winter from a VC standpoint... but even in the 90's there was some (GOF)AI stuff still going on after that.

There are too many actually useful things coming out of this for a true winter. And for there not to be a bubble.

They get there in completely different ways though. If there's no answer Prolog will fail out, and GPT* will usually make (stuff) up.

IMO it's "You don't have to use a Pi for that"

Also the later generation thin clients (i.e. Dell/Wyse 5060, 5070) are cheaper still and also pretty good. They're reasonably responsive for desktop tasks and can even drive 4k60 over DisplayPort.

BERT's more likely to be one than GPT. (Ernie-snicker ;) )

Also it's a complex adapter, since it has to be a PCI-e host. Generally SATA->x (USB, SD card, etc) adapters can be done with much simpler app-specific microcontrollers with IP blocks sprinkled in.