dylan (@dylanvsthe) header
dylan (@dylanvsthe) profile picture

@itsalltruffles. views my own

Canada
Joined November 2016
Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)Photo by dylan (@dylanvsthe)
If AI models are the 'new processor', why is everyone stuck doing single core? So we just did things for a bit (not without struggle) and shipped the first Orin SOM core that can parallel process inference requests to multiple models at a time. Thread inbound... 🧵 (1/N)
10
33
6
347
108,596
zen and the art of truffle maintenance
I grew up around people who fix and build things. Truffle is what I wish they'd always had. A personal agentic computer that does real work, so the people who actually make things can move faster.
1
18
527
you guys know you can just rewrite the kv cache right?
3
352
if i am hitting astra misalignment checker every 5 minutes is the model misaligned or am i
3
230
5.6 feels like if opus 4.8 wasn’t a coward
6
1,497
truffle is my favourite lab
8
641
does claude feel pleasure when the user prompt hits the J-spot?
Replying to @AnthropicAI
The J-space lets us read, audit, and shape what Claude is actively thinking about—useful tools for keeping models trustworthy as they grow more capable. And it suggests surprising parallels between language models and our own minds. Read the full paper: transformer-circuits.pub/202…
1
7
2,478
fable is very joycean
141
i have gotten into the habit of calling a remote access token the truffle uses the 'RAT' for short. my OpenAI account just got flagged on codex because of this (cyber abuse? wrong rat?!), and appeal was rejected, despite attempts to followup with more context once I pieced together the likely cause. @OpenAIDevs @thsottiaux do I seriously need to watch my acronyms or be banned from accessing your intelligence?
1
1
7
631
i haven't even used Codex in the last week (on vacation) and yet I got another warning, no response from appeal, and am now banned. great. @jxnlco @reach_vb i really liked being an openAI customer? i am at a loss for words
1
1
2
253
scared to write an appeal to the ban, knowing I only get one shot to defend, and my last few appeals/guesses clearly weren't right. my work is lowlevel/systems/authy - but shouldn't be going near cyber? I swear it must be auto-reviewing / memories and getting me banned repeatedly?
2
172
airline copilot "controls are yours" but it's me and codex fighting over my GPU
134
With the latest update, Truffle has access to; Amazon Postmates Spotify Uber Google accounts Outlook Resy & Opentable Socials (X & Instagram) V0.2-alpha out this weekend
16
6
6
269
20,381
> ai generated content is going to ruin the world > tries sora 2 > the slop has me in it > wtf i love slop now
2
5
657
i tend to find ai content does give you a form of temporary stockholm syndrome. never fairs well against the test of time. curious to see how sora hitting the masses plays out
2
4
457
excited to be in SF this weekend helping judge this event... looking forward to seeing what people build!
Introducing: The Voice of AGI We're hosting a voice agents invention and building party where hackers will launch AI Avatars, dating interview agents, and Zoom integrated voice tools📢
2
4
1,005
i've actually never been to the bay area.. so beyond looking at the bridge what else should I check out? down to grab a coffee if anyone's around.
5
268