Dimitri Zhorzholiani (@dimitrizho) header
Dimitri Zhorzholiani (@dimitrizho) profile picture

Co-founder & CEO of https://teapaw.org/t.co/AR7dHYeUNE | Co-founder & CTO of @Pepzilla_org | Writing about tech, fitness, and innovation.

Tbilisi, Georgia
Joined November 2023
Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)Photo by Dimitri Zhorzholiani (@dimitrizho)
AI passed the Turing Test! Follow for the latest AI updates and insights: x.com/dimitrizho
15
35
11
392
695,342
Dimitri Zhorzholiani retweeted
if this PHP vulnerability allows remote code execution, doesn't that mean we can use the vulnerability to auto-patch all vulnerable servers?
383
1,743
498
21,837
1,371,196
Dimitri Zhorzholiani retweeted
Ilya was right and predicted much of this
118
189
27
5,246
565,205
Dimitri Zhorzholiani retweeted
Man what the fuck
135
77
11
4,331
331,365
The top open models on are now all by Chinese companies
130
GLM-4.6 outperforms claude-4-5-sonnet while being ~8x cheaper.
2
322
LACT "indirect undervolt & OC" method beats `nvidia-smi -pl 400` on 3090TI FE.
225
OpenAI usage breakdown released
61
Unsloth Dynamic GGUFs - Aider Polyglot Benchmarks
69
GLM-4.5 is now leading the Berkeley Function-Calling Leaderboard V4, Beating Opus 4
83
LLM speedup breakthrough? 53x faster generation and 6x prefilling from NVIDIA
1
86
The "speedup", is only at large context. they demonstrate lack of loss of performance, not gain of performance. They developed framework to replace in optimal way subset of attention layers in existing models with linear layers, as this suibset in non-trivial to find, and just slapping linear every say 3 layers instead of attention will degrade performance.
44
Qwen Wan2.2-S2V is coming soon Alibaba is doing his Alibaba things!
1
212
OpenAI, I don't feel SAFE ENOUGH
58
Anthropic’s New Research: Giving AI More "Thinking Time" Can Actually Make It Worse
1
126
Their paper, “Inverse Scaling in Test-Time Compute,” reveals a surprising phenomenon: in certain tasks, models like Claude and OpenAI's GPT-o series actually perform worse when allowed to "reason" for longer.
80
Qwen3-Coder is released! This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation
1
79
UTCP: A safer, scalable tool-calling alternative to MCP
4
197
Kimi K2 tops creative writing benchmark
1
176
Moonshot AI just made their moonshot
1
104