fubarx

AI software can't technically kill you unless:

It convinces someone who can kill you, to you know, kill you.
It gets access to the physical world, which can definitely kill you.

Oh, wait.

U.S. military is using AI to help plan Iran air attacks: https://www.nbcnews.com/tech/tech-news/us-military-using-ai-help-plan-iran-air-attacks-sources-say-lawmakers-rcna262150
Anthropic Previewing the Model Hardware Standard: https://www.anthropic.com/news/model-hardware-standard-research-preview

See, nothing to worry about.

Opt for a stretchy, elastic string. Your future self, racing to make it to the bathroom will thank you.

Anything Monty Python. Also, Gravity's Rainbow.

A as in Aight
B as in Bye
C as in Cite
D as in Deign
E as in Eight
F as in Foto
G as Ghee
H as in Hale
I as in Ire
J as Jose
K as in Kneel
L as in Llama
M as in Mow
N as in Nigh
O as in Ohm
P as Phone
Q as Quoin
R as in Row
S as in Site
T as in Thine
U as in Ute
V as View
W as in Woo
X as in Xerxes
Y as in Yow
Z as in Zillow

Another conspiracy hidden by Big Urine.

Everyone get on the same bus.

Long time ago, I had an indoor ginger cat. Rescue kitten. My girlfriend at the time kept teasing it, rubbing his belly, rough-housing with it. I kept telling her to go easy, but she thought it was fun.

Pretty soon, the cat learned about indiscriminate biting and scratching. All my family and friends hated being around him. I had to lock him up in another room when people showed up.

The girlfriend got a job out of town and moved out and I had to move to a place that didn't allow animals. Fortunately, I found a friend with a farm who wanted to take him. Last I heard, he was having the time of his life riding on top of sheep and horses.

Humans encounter a wide range of tasks as they grow up from infant to toddler and beyond. We don't have separate 'training' vs. 'inference' stages (or at least, it's a very short window). We also generalize from tasks (as the glass jar demo shows). We also appear to have different types of storage for short vs medium vs long-range memory.

We can combine a bunch of individual tasks into complex, coordinated, goal-seeking tasks and keep them around for later use. Like riding a bike, or learning how to drive, or juggle. We can recall those years later, even if we haven't done them for a long time.

Plus, we somehow have a meta-layer of 'common-sense' or 'ethics' that tells us when NOT to do something, or when we SHOULDN'T. Like ride a bike onto pedestrians, or juggle kitchen knives. Hopefully, the research will go into those areas as well.

It was a 'cult revival' a few years ago. The only movie I ever walked out of. My buddy stayed and says he enjoyed it:

Kull the Conqueror: https://en.wikipedia.org/wiki/Kull_the_Conqueror

This guy fingerprints.

(hopefully in a good way)

The WebAudio test is not the only measurement in these scripts. Inspection of the bundles found code that queries or measures:
canvas rendering and toDataURL()
WebGL renderer information, extensions, and shader precision
audio oscillator and analyser output
screen and viewport dimensions
device pixel ratio
hardware concurrency and device memory
installed browser plugins
supported audio and video formats
WebRTC behaviour
browser performance timing
mouse, touch, focus, and scroll events
device motion and orientation
properties commonly associated with browser automation
There is also code for serialising and encrypting results, making requests to Alibaba telemetry services, and sending data with fetch() or sendBeacon().

Not sure you need all this to play audio.

Wow. Totally missed the Qualcomm acquisition.

Mojo is a pretty ambitious project. But they need to figure out how they fit into the latest, fast-moving pipelines. The one niche they would be perfect for is local, high-speed, on-device inference.

No big. Guess I'll have to wait another year before upgrading my laptop. Sorry for the loss of business.

On a DGX with 128G unified I could push Qwen3.6-27B to 8-bit and it had pretty decent response. It just croaked with MLX on the Mac.

Waiting to see what Apple announces next month. Judging by RAM prices, it'll be waaay outside my budget.

Qwen3.6 with 4-bit quant was hallucinating HARD. When asked to summarize the plot of obscure, cheesy 80s books, it confidently made up plot and characters. Not a single statement was correct. It was glorious.

Any higher quant and it would take 30-90 seconds before first token. Both MLX and GGUF on an M3Max with 128G combined RAM. Will have to try it with this one.

If I had to do it again, I'd start with the end goal and work backwards. It was a long time ago, I was single with no extra expenses, had a decent job, and thought it might be fun to learn. If I had wanted to become a professional pilot, the calculation would have been very different.

Getting the cert mainly involved putting in the hours. Having a good instructor helps a lot. Also, not making any stupid mistakes. I almost stalled the plane hard on my first solo touch and go. The hardest part was getting used to the ATC lingo and not stuttering too much. I got an aircraft band radio with a headphone and joined all the weirdos who liked to sit at the end of airport runways.