iiinova — where your digital self keeps evolving

the blocking shall continue until the blocking continues?

0980

I low key think this account is some weird projection stuff. Like I’m wondering if it’s an attempted follow farm that went bad

3270

device in hands, ruled illegitimate

2630

they made this for you

2461

it doesn't even swap the power button's functionality with the mode selector if you enable the single-hand option, it just makes the power switch go dead. mystifying

080

actual step two: go home for a second because I forgot to bring a lens cloth and it spitty out here

3250

step one: buy ponchos, in case it gets more gross than expected step two, as everyone knows: forget ponchos in car later

4450

I mean, I use LLMs where they're good. it is never to make expert-level decisions. I think AI video 1) sucks, and 2) people who use it for disinformative or conspiracy-theory purposes should be in jail this is what I'm packing right now to go outside, btw, I do my own stunts

2260

I shan’t risk it

140

meant this to lead into #2, because we have to have little crazy before big crazy

070

that said some of their other claims are. uh.

290

this is a direct insult on multiple levels 1) what's this euro can nonsense?? 2) WHITE MONSTER?????

360

Ed @ed3d.net · 24/09/2026

I cannot be restrained merely by choice of platform but this was way harder than I thought it'd be

4251

but could it run doom

020

but I believe in practicals above all else, which is why my next video required touching some grass and smelling some ocean

0130

have you touched grass today ferns also acceptable

2252

enforcing PVP in channel comments

5811

Ed @ed3d.net · 22/09/2026

Here's an example: it describes a classifier: it sees X, it resolves to A. But when I described in the initial spec that I wanted something it hadn't seen in its training data before, it tried to do...this? Why is the conveyor splitting? Why are the question marks outside the screen? it's just dumb.

1190

I put reasonable levels of the frontier model graphs on here, plus GLM (they do max and GLM recs high, so not perfect) Ignore the vaunted green quadrant - anything over 40 is generally fit for purpose for tasks, and orchestration capability/adherence seems to mostly be model size cost is log scale

2234

Ed @ed3d.net · 22/09/2026

This is where benchmarks are going to start making progressively less sense in general. Like, on a pure happy-happy-joy-joy code-grind basis, 95% of developers might do something that would represent the top half of DeepSWE...what, once every five years? We have saturated HumanDevBench thoroughly

4484

updating my priors (goddammit why will this not leave my brain)

1311

four stars: did not violate physics

3261

she will not forget this betrayal

2170

she has decided that if she's going to be miserable we're all going to be miserable

2250