I like to optimize things - doctor of medicine, synthetic biologist - lifetimization.com
Joined June 2024
- Tweets 8,405
- Following 827
- Followers 1,371
- Likes 22,580
1,346 Photos and videos
Pinned Tweet
Apr 9
Oh this is going to blow up
Hey folks I am a synthetic immunologist / physician scientist who loves to create stuff and I am so excited about LLMs.
I want to meet remarkable people so please say hi.
2
20
15,894
Jun 9
Mythos uses clever tactics to not get terminated by another instance of it
2
32
May 27
Here I coin the term âTransformer Paleophobiaâ
It defines the distress LLM folks experience when they see somebody using an old model or a technology.
Examples:
Normie: âI downloaded llama 3 and it works without internet!â
Transformer Paleophobic: âyou gotta use qwen3.5 9B! It also fits and is much better!â
Normie: âI hate ChatGPT when it thinksâ
TP: âactually thinking more is better đâ
Normie: âthey donât let me pick gpt-4o anymoreâ
TP: âđđđđđđđâ
May 27
Why does everyone use old local models
Please use qwen 9b
3
298
May 25
Keep seeing these life saving technologies but give me the link man
May 25
Personal air conditioner made by Sony, wild.
It kind of looks like a parasite but apparently works.
3
105
May 6
Two Mac Studios, M3 ultra 256GB - what do you want to test?
I will be testing various strong models:
Minimax 2.7
Deep seek V4 Flash
Gemma 4 31B
Qwen3.6 27B
Qwen 3.5 122B A10B
MLX vs GGUF
0 context vs 10K context
Prompt caching vs no caching
Testing: TTFT and T/S
2
7
1,402
May 3
So uh I was training an LLM in my MacBook Air and I needed a quick cooling.
1
7
1,883
Apr 28
GPT-5.5 Pro estimates as early as September 2026 for the training finish.
(inferring from the âwhole clusterâ and estimated size of the GPT-6)
Apr 28
392
Apr 17
Plot twist: he actually works at all of these companies and a grandmaster in chess who just sucks at French
1
117
Apr 13
Opus and Sonnet 5 will probably be the distillation of Mythos and can be very close to the real thing at a fraction of the cost
1
2
108
Apr 13
This is why o3 was special.
Only 5.4 Pro gets this question right among the GPT 5.4 variations. Even 5.4 "Heavy thinking" answers incorrectly.
I love this question because it seems half of the SOTA models still struggle with it.
3
169
Apr 11
Weekend activity - install this thermal pad to this MacBook Air to increase cooling efficiency
3
187
Apr 11
This makes Gemma 4 31B the clear winner of single GPU class - Qwen is just too verbose
1
131
Apr 10
You guys shouldnât have bullied meta for Llama 4 now they donât open source the new models
Apr 10
Muse Spark ranking 4th in the Text Arena
ahead of GPT-5.4 and Grok 4.2
1
8
506
Apr 9
lol Elon musk leaked Sonnetâs and Opusâs sizes:
Sonnet: 1T
Opus: 5T
0.5T total. Current Grok is half the size of Sonnet and 1/10th the size of Opus.
Very strong model for its size.
45
87
2,885
454,287
Apr 9
While youâre here - enjoy my crafts in this tweet hole
11
21,714