Sidney VanNess (@sidneyvanness) header
Sidney VanNess (@sidneyvanness) profile picture

Founder On Call Central & Recursive Capital. Interest in tech, finance, aviation, genetics & neuroscience. I build new things.

Granite Bay, California
Joined November 2007
Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)Photo by Sidney VanNess (@sidneyvanness)
North Korea: most successful entrepreneur of the AI era. Achieved $6.3B in revenue in about 2 hours.
26
Movement in the AI model Pareto frontier over the last 29 months is incredible. I decided to collect and visualize the data. Here's the entire frontier over the past 2+ years. github.com/sidneyvanness/che…
1
1
2
412
GLM 5.3 Flash, which vastly exceeds anything that was available a short while ago, is roughly 1/125th the cost of the most capable model at the beginning of this data set. Seems like we are still early and have potentially a few more orders of magnitude more efficiency to gain.
1
50
Internal dialog of GLM 5.3 Flash, reasoning set to medium.
“The WIFI password is on the back of the router” The back of the router:
2
259
It's DNS. It's always DNS. Probably DNS in a common data center used by all.
53
This is dangerous territory. Highly capable abliterated model, explicitly marketed for offensive capabilities. Zero data retention. Little prevents a bad actor from using this to anonymously PoC major attacks against high value cyber targets at scale.
Today we're releasing abliterated-model-large-v2. Based on GLM-5.3, which is #3 on Terminal-Bench 4.0 (behind only Opus 5 and Fable), with 2× the cyber exploitation of 5.2. We abliterated and hosted it so it does the offensive cyber, red teaming, and agent testing work other models refuse to do. - US-hosted - FP8 - 1 million context window - Zero input/output prompt retention Live now. 🧵
1
2
113
The DeepSeek V4 Flash Vision model + Hermes Agent, I am finding, is far faster than running GLM 5.3 Flash on the same setup.
1
2
187
Hard to see how @AnthropicAI continues to limit Fable use on subscribers when @deepseek_ai v4 Pro is less than 1/50th cost on metered use, functionally equivalent, and can be run with zero data retention. Why would anyone deal with "your limit resets next week"?
1
113
Current frontier AI model’s favorite word is "shape." Shape of the data. Shape of the workflow.
105
That was maybe the weirdest @nbcsnl in a long time.
2
163
"Directionally" is the word du jour of every frontier LLM.
2
147
Looks like Apple completely removed the 512GB RAM option for the Mac Studio M3 Ultra.
2
219
If you're mad about @openclaw or @steipete joining OpenAI, just fork their repo. It's MIT. About as permissive as possible.
76
Memorable day flying N464WH to/from Monterey.
3
97
If you can't get $200/mo out of @claudeai, Claude isn't the problem. You just don't have big enough ideas. Absolutely incredible.
1
82
Sidney VanNess retweeted
Has it honestly never dawned on the EF that OGs might help/contribute more if they didn't treat us like yesterdays dogshit? EF has a serious respect problem. @peter_szilagyi @nicksdjohnson @VitalikButerin @adietrichs @hudsonjameson @avsa @d11e9 @simondlr @jeffscottward
2
5
2
17
1,538
Auburn, CA to Albuquerque.
3
1
10
271
I’ve managed to put about 30 hours on N13180 in the last month. Trips to Wyoming, Utah, various places in California. So much fun seeing this beautiful country from this perspective, and with friends and family.
2
5
238
Managed to get in a bunch of flying this week with friends and family. First with my daughter, then with my wife, then my buddy Adam. It’s fantastic to be able to share this experience with people you love!
1
3
1,079