@DeeperThrill

🤖 Entrepreneur. Healthcare AI. Biomedical Engineer PhD. Futurist. Optimist. Unvaxxed. 39 yo.

t.me/deepthrill
Joined December 2013
  • Tweets145K
  • Following6K
  • Followers37.1K
9.4KMedia
Lmao just don’t sneak into countries illegally or illegally overstay your visa. It’s not complicated, and they know what they’re doing. These aren’t “oopsies”, they’re “I hope I don’t get caught”.
19
821
Deep Thrill retweeted
Chris Munns
@chrismunns22h
In a somewhat normal world we’d be hauling the CEOs of major TV companies in front of Congress to ask why they’ve built this shit. The answer can easily be Five Eyes told them to, but we can cross that bridge later.
mike
@mike64_tOct 25
Ok, I just checked this, and its literally running a freaking TensorFlow Lite model on 960×540 images with INT8 quantization every couple of seconds in an attempt to figure out what game you are playing. It's literally called "alex.tflite" and its literally AlexNet and seems to follow torchvision.models.alexnet. As far as I can tell this does some strange nearest neighbor feature comparisons against a DB and literally doesn't work 99% of the time, you love to see your CPU get wasted every other second by this thing though. It also just seems to dump its results into the void because the consent gate comes later. Anyways, here's the tf model graph and some dumped captures.
15
116
1
1,377
18,136
This is wild. Change a few neurons out of billions and you get sinister model! And this is how it was found.
lyra
@_lyraaaa_Oct 25
Replying to @Orwelian84 @adhvfed
i asked claude very nicely real answer tho- logitlens, steerlens, and mapping neuron influence maps across the entire model, then text search + reading largest activations on input text -> filter by checking what it pushes and steer to verify. about 2-5 mins per neuron search
1
5
1,076
Go ask your favorite agent right now to reduce the number of lines of code in your repository by 50%. You’ll thank me later.
5
1,117
Deep Thrill retweeted
Max
@minordissentOct 23
the reason you as a 130+ IQ chud need to become rich and powerful is not because it will cure your depression (though it will) but because it will cure your interminable irritation at the law of entropy. No longer will you have to be inconvenienced by things always breaking or retarded people being retarded. You can just throw money at the problem by replacing it or hiring someone competent to handle it on your behalf.
127
604
51
10,868
689,209
Speed is the new intelligence. Tokens per second is what matters now.
Fireworks
@FireworksAI_HQOct 21
DeepSeek V4.1 Flash 1017 runs at 1,521+ tok/s on Fireworks Ultrafast. Fast enough to ship Tetris in 11 seconds, 7x faster than Claude Opus 5.5. Join the Ultrafast waitlist: bit.ly/3UA0wKq
6
754
Deep Thrill retweeted
Justine Moore
@venturetwinsOct 21
Sydney Sweeney reveals in an interview that she doesn’t care about the PHP vulnerability because she moved everything to Rust in 2024. “They said the rewrite would take six months. We’re on month 30. We haven’t shipped a feature since, but I’m having a very normal Wednesday.”
264
1,714
142
27,846
709,235
Speed is the new intelligence. Instead of benchmark-maxing, model inference providers better start tokens/second maxing.
0xSero
@0xSeroOct 22
We are hitting levels of speed you can't imagine. Deepseek-v4.1-flash-1017 @ 675.9 tok/s c1 Recipes coming
8
755
I often have Hermes coordinate parallel cli runs of OhMyPi and OpenCode with OpenRouter using either DeepSeek or GLM. What’s your harness setup look like for coding work?
2
666
LFG, more attacks means smarter defense! Also it’s wild what AI can do, loving seeing its effects in the wild, finally.
The Hacker News
@TheHackersNewsOct 21
🚨 CRITICAL: GitLab flaw (CVSS 10.0) affects 12 years of releases. CVE-2026-110264 could allow unauthenticated remote code execution on self-managed servers. Read - thehackernews.com/2026/10/cr…

Critical GitLab Flaw (CVSS 10.0) Could Allow Unauthenticated RCE Across 12 Years of Releases

GitLab fixes CVE-2026-110264, a critical flaw that could allow unauthenticated code execution on self-managed versions dating back to 7.0.

thehackernews.com
2
1
10
890
This is actually healthy to harden the digital infrastructure.
Varunram Ganesh
@varunramOct 21
Railway is compromised, rotate all API keys as soon as possible Deployments are down so don’t wait to redeploy Attackers are draining keys in real time, we didn’t lose much but can confirm this is still an active attack Stay safe friends
1
9
999
“Yo I told this homie, ‘Stop sending me AI animations! Haha I can’t even!’ Like motherf*cker my sides can’t take it no more!” -overheard at a local bar. To imagine that “AI” would enter this vernacular of day-drinking locals even 3 years ago woulda made us scoff with disbelief
1
5
635
I just told GPT SOL 6.1 High to fix 5 open GitHub issues on my repo. It is 4.5 hours in to that single prompt, and not yet done. Would it really have taken me an hour to fix each issue? I’m not convinced it’s actually faster than me. I guess it frees me up to do other stuff, and it can scale in parallel, but any claim that AI is actually a faster or better coder than me on an individual task is still very suspect.
4
9
962
This is ridiculous. A single prompt that said to fix 5 open issue. 6.5 hours later, GPT 6.1 Sol High is still churning. I coulda had this all done in 90 minutes if I was sufficiently caffeinated…
110
The problem with letting the machines write all the code is sometimes you read it and go, “ugh, it’s doing it wrong!” Solution: just don’t read the code the agents write?
3
10
1,061
Opus 5.5 is such a goated model. I like it much better than Fable 5.1 and Astra 6.
4
12
902
Do you ever just prompt an agent and then sit back and think to yourself, “Damn that was a great prompt I just wrote!”
2
9
982
Context management is so important. You have a million tokens to work with. Deciding what that context window gets filled with, in what order, on each stateless LLM invocation, is crucial to get right. The following are some components of each LLM call. What percentage of context tokens allocated to each of these, and what order they appear, is important to think about. 1. System prompt(s) 2. soul or agents always-loaded markdowns 3. List of tools & skills with short descriptions for each 4. Compaction summary 5. Chat history 6. Tool call results There’s more, but those are some of the main pieces.
1
3
535
Deep Thrill retweeted
David Kobia
@dkobiaOct 20
Replying to @Austen
I don't think understanding the code is important. Understanding the architecture is. Bad boundaries, poor data flows and a code base that grows exponentially with each vibe code spin is not good for resilience.
2
1
14
869
Load more