apokryphos
@apokryphosx
Followers
29
Following
4K
Media
9
Statuses
70
Just finished my first ever stream on x, building a pipeline to create synthetic data for SecureSage. Thanks for joining me @wambosec.
1
0
2
Gemini 2.5 pro is insane compared to the watered down o3 holy shit.
0
0
2
How would you improve the capabilities of LM agents with terminals?.
0
0
1
Closed source is the devil.
new o3 is dumber actually. oai didn’t reduce the price out of charity… . they nerfed o3 . the new silent update has made context length shorter and the answers are dumber and “quick” responses. after speaking with researchers, i’m told openai believe to have a tiny improvement.
0
0
2
Time to clean up the tl, too many things I don't want to interact with.
0
0
1
"If you would please consult my self made graph, you can clearly see right *here* that I am right and you are wrong".
Time to calm down about AI, at least about LLMs. A rough visualisation of the "diminishing returns" view of current progress, as explained by @GaryMarcus
0
0
0
Can this app not be bugged the day after anthropic has released.
0
0
0
RT @vikhyatk: > new "open source" model released by google.> blog post about fancy new architecture.> no benchmarks, no inference code http….
0
2
0
Average SOTA LLM paper nowadays:. The Value function gives us the expected cumulative reward from state s_t following policy π.
2
0
1
My next 3-4 hours are sorted out.
❄️Introducing Absolute Zero Reasoner: Our reasoner learns to both propose tasks that maximize learnability and improve reasoning by solving them, entirely through self-play—with no external data! It overall outperforms other "zero" models in math & coding domains. 🧵 1/
0
0
3
Grok: Yes of course I can tell you the name of H.P. Lovecrafts cat. Its name was. GPT 4: I understand its frustrating that you're code isn't working. To address the error, we need to change the display. Here is the updated code: *same code*.
3.7 sonnet: *hands behind back* yes the tests do pass. why do you ask. what did you hear. 4o: yes you are Jesus Christ's brother. now go. Nanjing awaits. o3: Listen, sorry, I owe you a straight explanation. This was once revealed to me in a dream.
0
0
1
RT @Jush21e8: 99% of the code it writes (yes i tried gemini 2.5 pro), it gets it wrong or write so much unnecessary code and i just end up….
0
1
0
> mfw increasing batch size times 8 makes the run almost 3 times faster despite GPU utilization not moving an inch.
0
0
0
"Oh, this training run will take 2-3 hours. My GPU's are 100% utilized, even if there is still roughly 60% of the available VRAM left. That's a bummer, can't do anything" - me, a retard.
1
0
0
TIL just how much of a bottleneck I/O is when working with actual compute.
1
0
0
