solipsistic toaster

Anime accelerationist. ML and things

[solipsistic toaster @toasteratemycat]

@SaladinAlDronni One can wonder how much they overcharge for Yolka clone from 2 years ago

[solipsistic toaster @toasteratemycat]

@NotWoofers the idea might be to draw drone teams’ focus from infantry push that follows as recent info from such attacks suggests

[solipsistic toaster @toasteratemycat]

@Rebel44CZ systems that can jam those satellites are expensive and easy to hit with ballistics. even if range will be more than 20km, it still needs to be placed in every major city, military sites, airports etc With West’s infamous rate of production I wouldnt underestimate the issue

[solipsistic toaster @toasteratemycat]

jackpot for intelligence units with access to live satellite data

[Stefan Amberger @__snamber]

Tried this across an entire German state

Three prompts and Tilebox ran zero-shot Qwen3.8-Max across Brandenburg. We found 247 PV power plant candidates.

82 km² at high confidence, official capacity implies ~87 km².

Almost the whole expected footprint in one shot!

[solipsistic toaster @toasteratemycat]

@cheapncheerless there was no Kh-101 launched on that day, so either 9M723 or KN-23, leaning towards the later and afaik not a direct hit

[solipsistic toaster @toasteratemycat]

@cameron19460429 at this point experience and know-how of making FP-5 is as important as missile itself, so redesigning another one will go much better

[solipsistic toaster @toasteratemycat]

@roman_koshchei @dim0kq Ймовірно погано збалансований датасет на такі ракурси та клоузапи

[solipsistic toaster @toasteratemycat]

@roman_koshchei Duplication is a safe choice but afaik the algo was recently changed to boost posts within the same region/country no matter the language

[solipsistic toaster @toasteratemycat]

@RALee85 Decoy detection is kinda hard anyway, cause in some cases you can only spot differences up close when there is little time to pull the drone away from striking

[solipsistic toaster @toasteratemycat]

@RALee85 No indication present that it tricked the system to recognise it as AD, though. Not that familiar with Hornet UI, but perhaps it was selected for tracking by the operator as a "Manual" label is present on target

[solipsistic toaster @toasteratemycat]

@BlackHC The AI Act is a step in the described direction and an acceptable framework to build on. Though I'd argue that, at the moment, the greatest danger is posed by threat actors who can just use highly capable, non-regulated models from Deepseek/Qwen/Z.ai, etc.

[solipsistic toaster @toasteratemycat]

LoRA weight modulation for recursive DiT is quite good so far, hope that scales

[solipsistic toaster @toasteratemycat]

@fleetingbytes more of a graphic design art piece than technical illustration, I can see it just hanging on the wall or as a shirt print

[solipsistic toaster @toasteratemycat]

Bait used to be believable

[Amit LeVi @AmitLeViAI]

Such a great evening to start a brand new research for NeurIPS in 3.5 days.🧘‍♂️

Day 1: planning.

Night 1: running experiments and sending the abstract.

Day 2: reading results fighting with Claude, and sending again.

Night 2: sleep (optional).

Day 3: opening Codex, and finally, write the pape in parallel.
Night 3: resolving the “beef” with Claude (temporary peace) and going to sleep.

Day 4: final reading, last-minute fixes, submission then some relaxation, maybe a beach walk.

I’ll keep you posted on the results.

This will be my only single-author paper, so I can’t hide behind other submissions if it gets rejected 😅

[solipsistic toaster @toasteratemycat]

yep saw that coming
it's surprising that their unsustainable model even lasted that long
given the same moves from other providers, guess we'll soon learn what is the real price of the token

[GitHub @github]

Starting June 1st, GitHub Copilot will move to a usage-based billing model as GitHub Copilot supports more agentic and advanced workflows.

In early May, you'll see a preview bill experience, giving visibility into projected costs before the transition.

👉 Read more about the upcoming change: https://t.co/4IC9VNHwhk

[Macrophysiological System🐀 @InVitroFuture]

I find LLMs very helpful for scientific writing. I do the legwork of planning out what to say, track down citations, sketch out the flow, feed all of this to the LLM to generate a draft, and the output is so awful it motivates me to write it out the right way in disgust

[solipsistic toaster @toasteratemycat]

there's so much interesting papers from this year's ICLR and all at once, exciting stuff but might need a month to process all the ones I've saved

[solipsistic toaster @toasteratemycat]

going even further, its likely that the generative models of any modality learn representations that can be used for specialized tasks. Model capacity and data diversity certainly help. In some way it is similar to how LLM beat other approaches on classification or sentiment analysis, and while really useful to generate synthetic data and pseudolabels for distillation, straight up deploying the billion parameter monstrosity for downstream tasks surely is not the right call for most of the people

[Songyou Peng @songyoupeng]

Yay, finally! Introducing Vision Banana🍌 from @GoogleDeepMind, our unified model that outperforms SoTA specialist models on various vision tasks!

By treating 2D/3D vision tasks as image generation, we unlock a new foundation for CV.

Project page: https://t.co/GQgRi6mWwC

(1/5)