aespa(x)

1.7K posts

aespa(x)

@aespa__init

A certain scientific GPU accelerator. Work in tech but not technical. LLM enjoyer. Math challenged. Semi-serious. https://t.co/1H1CfMOsmO

Canada Katılım Şubat 2024

515 Takip Edilen80 Takipçiler

Sabitlenmiş Tweet

aespa(x)@aespa__init·3d

yann would never

English

1.6K

aespa(x)@aespa__init·3h

@HououinTyouma wow finally a real ball knower on my timeline

English

Mad ML scientist@HououinTyouma·20h

I have seen enough: GPT-5.6 Sol is AGI

English

1.3K

aespa(x)@aespa__init·3h

Since Grok didn't delete all the secrets on its own, it's safe to say it's an unaligned model? 🤔

English

aespa(x)@aespa__init·4h

xai continues to be unfriendly to enterprise customers I guess

English

aespa(x)@aespa__init·7h

@xlr8harder they seem to have fully recovered from the DoD fiasco and then some

English

xlr8harder@xlr8harder·7h

The bar for frontier labs to gain user trust is not particularly high. It's a shame OpenAI is the only lab that seems to care enough to try at all.

English

1.4K

aespa(x)@aespa__init·9h

Too dumb to be a researcher but too smart to larp as an AI influencer. Going to die a bum.

English

aespa(x)@aespa__init·20h

Tried to have Terra drive Adobe Lightroom and it was awful lol. Maybe try Sol tomorrow...

English

aespa(x)@aespa__init·22h

Seedance 2.0 tech report does reference it but how it's measured is not revealed.

English

aespa(x)@aespa__init·22h

There doesn't seem to be any benchmarks on cinematography except for CineTechBench, and that's over a year old... identifying the camera technique is a good initial step. AI videos tend to exhibit poor editing, which, to be fair, is emblematic of many popular mainstream movies.

English

aespa(x)@aespa__init·1d

@lithos_graphein limit up tomorrow

English

🌿 lithos@lithos_graphein·1d

Everyone's X feed rn.

English

2.8K

aespa(x)@aespa__init·1d

I don't know how people do live commentary of system cards since they're 200+ pages... Anyway, I am going through the Mythos one and this seems to say Mythos prefers not seeing the "offending" requests (i.e., routing to Opus 4.8) over Anthropic doing runtime changes...

English

200

aespa(x)@aespa__init·1d

If my resume is rejected by Fable, does that mean I should be hired or fired?

English

aespa(x)@aespa__init·1d

A bunch of words not found in the bible.

English

aespa(x)@aespa__init·1d

@lu_sichu x.com/drewpusateri/s…

Drew Pusateri@drewpusateri

Our statement in response to this suit: We have no interest in other companies' trade secrets. We remain focused on building innovative technology that empowers people everywhere.

QME

Sichu Lu@lu_sichu·1d

has anyone at openai responded to these claims, what's their side of the story exactly

Jack@tracewoodgrains

Wow. This—does not look good for OpenAI.

English

2.4K

aespa(x)@aespa__init·1d

What are people using GPT5.3-spark for these days? Maybe I should offload some web scrape scripts here to balance out my usage...

English

aespa(x)@aespa__init·1d

@N8Programs their cots are all like this

English

N8 Programs@N8Programs·1d

@aespa__init qwens haven't passed it for a while.

English

148

aespa(x)@aespa__init·1d

Danganronpa bench

N8 Programs@N8Programs

As a concrete example, Qwen, when asked about the '16 characters in Danganronpa V3' - a Japanese visual novel - goes into an infinite loop trying to recall the 16 names.

Indonesia

211

aespa(x)@aespa__init·2d

Someone will show a slide/diagram and say something. Someone will point out a mistake. "Oh yea haha I made this with AI." This happens more and more every week.

English

aespa(x)@aespa__init·2d

In Mononoke Hime, the first Ghibli film that employed digital painting and CG, the artists retouched the images to make them "look more hand-drawn".

English

aespa(x)@aespa__init·2d

I used to think future work will converge towards running on local models (to save token cost), and then you will just load LoRAs depending on the task, but the current cohort of white collar workers can't make anything beyond a html report, let alone useful agents.

English