Matt

2.7K posts

Matt

Matt

@Matthewagi

agi is here but it got an email job

Katılım Ağustos 2025
997 Takip Edilen201 Takipçiler
Matt
Matt@Matthewagi·
Does your dishwasher also have that smell? I'm fairly certain most people have a higher detection limit and it's harmless but the wet dog smell/taste is real and likely from dishwasher or water source. My fiance can't smell it but I can and sometimes need to rewash things from the dishwasher
English
0
0
0
41
2HP goblin advisor
2HP goblin advisor@goblinodds·
sense of taste got super sensitive all of a sudden and now tap water tastes like wet dog smells and i cant stomach meat F
English
21
0
131
7K
Matt
Matt@Matthewagi·
@SinaHartung But what about the hantavirus virus virus?
English
0
0
0
83
Sina
Sina@SinaHartung·
the hantavirus virus can’t get you if you’re lock in building ai agents and never leave your house FYI
English
6
0
15
564
Matt
Matt@Matthewagi·
I don't think so considering the body dumps 10x (100x for pro cyclists) routinely even when it's primarily generated in the legs. Only issue might be flow density but you could probably dump 50% more without issue but that's a large proportional energy drain which goes back to your main point
English
0
0
0
71
Matt
Matt@Matthewagi·
@ashebytes This is good advice for most things. They kill when there's a good template
English
0
0
0
6
Matt
Matt@Matthewagi·
@emollick I think a little hype still but that's a good omen
English
0
0
0
216
Ethan Mollick
Ethan Mollick@emollick·
So Mythos was, indeed, not marketing hype. Remember this is a general purpose model that just happens to be good at finding exploits because good models are good at lots of things. Expect similar from OpenAI & Google. And from open models in 8 months. hacks.mozilla.org/2026/05/behind…
Ethan Mollick tweet mediaEthan Mollick tweet media
English
79
130
1.6K
178.3K
Matt
Matt@Matthewagi·
wasting a day to see if the cheat way will work then implementing the direct thing that will work
English
0
0
2
33
Matt
Matt@Matthewagi·
@Duderichy Oh no I'm having sex with a chat bot
English
0
0
1
78
Matt
Matt@Matthewagi·
@gleech its called foreplay
English
0
0
1
17
Matt
Matt@Matthewagi·
@g0da_s You can probably add the efficacy of sleep trackers to make this stronger (most trackers are really really bad)
English
1
0
2
134
Goda
Goda@g0da_s·
sleep trackers are useless. here's what we built instead.
English
44
17
198
13.5K
Nam Huynh
Nam Huynh@TheNamHuynh·
@Matthewagi audience research with synthetic personalties. DM me for early beta access
English
1
0
1
9
Nam Huynh
Nam Huynh@TheNamHuynh·
ManyMe AI now has a Preview of Oracles that allow you to compare more than one LLM concurrently. Also more models are now available 🧵
English
2
0
1
38
Matt
Matt@Matthewagi·
@tokenbender love it and this would change the training paradigm again. engineering around pruning and distilling get really interesting
English
0
0
1
27
tokenbender
tokenbender@tokenbender·
@Matthewagi yeah, think being able to extract agentic/tool use abilities fully shaped by big model training but reusable at an OOM lower cost. also, i may be emboldened to think meta capabilities are discoverable and usable?
English
1
0
5
212
tokenbender
tokenbender@tokenbender·
Ever wondered if you could extract capabilities and behaviors from neural networks and reuse/update/route it as needed? We introduce low-rank circuit conditioning, a novel approach that preserves the model's output behavior while reshaping how an existing capability is represented. In the base model, standard compact recovery stalls at 29%. After conditioning, the same extraction pipeline reaches 91.33% autoregressive full-answer recovery from 5.05% of MLP channels. The evidence points to a possibility of extracting and using isolated capabilities saving cost, latency and high adaptability. Read our work to understand more - tokenbender.com/posts/honey-i-…
English
24
61
335
30.3K
Matt
Matt@Matthewagi·
@creet_z Ahem people? That's AI land now
English
0
0
0
56
Matt
Matt@Matthewagi·
@thdxr They suppressed horny Claude and this is how they manifest
English
0
0
0
407
dax
dax@thdxr·
call me claude the way i go down every day
English
37
20
926
39.7K