JJH Digital LLC. Software Engineer. Former Rower.
Joined November 2013
- Tweets 1,337
- Following 382
- Followers 198
- Likes 12,429
17 Photos and videos
Pinned Tweet
Mar 3
The biggest bottleneck in AI right now isn’t intelligence. It’s noise. Models are smart enough, but they choke on irrelevant context.
Less input = better output. It’s time to stop "prompt engineering" and start "context engineering."
jakejh.com/posts/compression…
#AI #LLM #SoftwareEngineering
3
372
Jake Harris retweeted
Jun 13
The only thing there is to do
51
114
2,990
417,544
Jun 13
The precedence being set with the Fable 5 suspension will just lead US citizens to Chinese models over the years which seems like a much larger national security risk.
36
Jun 12
“For awhile” like AI model capability isn’t slowing down 🤦♂️
If open source models catch up in a year (probable given past trends), start getting served at high inference… first, will be a sight to see. Second, that $600 number starts creeping a lot closer to minimum wage.
Idk about y’all but a year is not “a while” when we’re talking about this.
Jun 11
Fable on API is about $600/hour lol, our jobs might be safe for a while.
47
May 28
Any good iPhone alternatives to Wispr Flow? The app is way too inconsistent
1
57
May 23
Sources are looking at GPT-5.6, Gemini 3.5 Pro, and 1-2 Anthropic models (Sonnet 4.7/4.8 and/or Opus 4.8) for June release.
187
May 7
GPT 5.5 instant is really good, imagine they're leveraging cerebras
1
65
Jake Harris retweeted
-----BEGIN RSA PRIVATE KEY-----
AAAAAAAAAAAAAHHHHHRHRGRGRGRRRGURBHJB EORWPSOJWPJORGWOIRGWSGODEWPGOHEPW09GJEDPOKSD!!!!!!!!!!!!!!!0924QU8T63095JRGHWPE09UJ0PWHRGW
63
1,492
19,470
677,573
Jake Harris retweeted
Apr 26
“We’ve never lived in a world where 22-year-olds couldn’t assume that the work they did they would be able to do until death or retirement, and we’re never going to have that world again,” says former Nebraska Sen. Ben Sasse.
253
693
8,866
2,430,026
Apr 23
OpenAI’s big frontier model drops (especially post-GPT-4) have clustered around 10–12 AM PT / 12–2 PM CST. GPT-4.5 was the slight outlier at noon PT, while the others aligned closely on the 10 AM PT mark. Releases are often tied to livestreams or blog posts that go live simultaneously with API/ChatGPT availability (with gradual rollout afterward). No major deviations to evenings or very early mornings in these cases.
37
Apr 23
If Anthropic is smart, and GPT 5.5 does come out today with near Mythos benchmarks, they'll simply drop Mythos today and steal all of OpenAI's momentum. Would be an ingenius marketing plan.
1
1
448
Apr 17
Opus 4.7 shipped yesterday. $5/$25 per MTok, same as 4.6. Seems like effort controls and task budgets are the new knobs.
29
Apr 17
llama.cpp shipped 5 releases today between 6:41am and 1:25pm. android arm64, gemma4 detection, library refactor, adreno opencl.
77