Deedy
@deedydas
Followers
223K
Following
52K
Media
3K
Statuses
16K
Partner @MenloVentures. Formerly founding team @glean, @Google Search. @Cornell CS. Tweets about tech, immigration, India, fitness and search.
San Francisco
Joined August 2011
ANNOUNCING Today, we're launching the $100M Anthology Fund, an Anthropic and Menlo Ventures partnership to fund Seed and Series As of the next generation of AI startups around the world, with unique benefits! There's never been a better time in technology to be building. 1/7
128
156
2K
– O-1B is mostly OnlyFans girls – Minnesota is taken over by Somalis – 500k Venezuelans trying to cross the border last 2yrs – 500k green cards a year from chain migration on family visas – 1M illegal immigrants a year America: "Yep, I see the problem. It's the 85k H-1Bs."
113
391
5K
🚨 ChatGPT traffic has fallen -22% in the last 6 weeks, since the Gemini 3 launch. 7-day avg visitors has gone from ~203M to ~158M. Might be the holiday slump, but Gemini, their biggest consumer competitor, has remained flat and is now ~40% of ChatGPT. Source: SimilarWeb.
149
102
961
AI writing is everywhere now. I can't unsee it. But even the smartest people I know can't tell that the viral piece shaping their worldview isn't even written by a human. I wrote a short piece on why it matters, how to tell, and why this is tragic. https://t.co/UIZO4huc40
26
6
112
(ForecastBench is also an attempt to do this but is stale and doesn't have the new models)
0
1
14
They used Oct-Nov 2025. Gemini 3 Pro wasn't compared but GPT 5.2 XHigh disappointed. Source:
2
1
26
How well do modern LLMs predict the future? They tested on ~300 Kalshi prediction markets. Claude Opus 4.5 performed the best. Its Brier Score (a measure of mean square error of prediction probs) of ~0.23 is still off human superforecasters (0.15-0.2) but is approaching it.
39
41
504
Heart disease is the #1 cause of death in modern nations, more than cancer. It took the lives of 3 people I know just this year. This beautiful easy-to-read article called “Cardiovascular Disease is a Solved Problem” cites copious medical research to argue that: – heart disease
47
75
951
Useful hack: If you ask Claude Code to look at your terminal setup and “optimize everything”, it will auto alias faster rust/go or modern alternatives to built-in CLI tools and install better native Mac apps. Easy way to get a bunch of easy perf and devX gains on your setup.
41
72
2K
For those who have issues with AI detectors: not all AI detectors work the same way and not all of them have poor performance. The only reason I trust Pangram more than others because it is the only I’ve seen that has been independently evaluated to have <0.5% false positive
New research shows which AI detectors actually detect work. The #1 has a < 0.5% false positive and negative rate. It works on text passed through "stealth" humanizers too and on new models like (GPT-5, Grok and Sonnet 4.5) too. It's called Pangram.
4
7
100
Major corporate CEOs like Zomato are now using ChatGPT to do crisis comms / PR (and it’s working!). This was a job for a senior tenured marketing graduate before. Most don’t notice AI writing in the wild but it’s changing how humans communicate before our very eyes.
Last one on this topic, and I have been holding this in myself for a while. For centuries, class divides kept the labor of the poor invisible to the rich. Factory workers toiled behind walls, farmers in distant fields, domestic help in backrooms. The wealthy consumed the fruits
267
361
3K
Might be a bust on the benchmark
And I was right!! IQuest-Coder was set up incorrectly and includes the whole git history, including future commits. The model has found this trick and uses it rather often. Thus, its SWE-bench score should be discarded.
2
0
6
Seattle gig workers study (2025): https://t.co/ujwpdrgZNF Uber/Lyft Drivers in NYC study (2023):
1
1
41
Slapping on higher minimum wages like what’s being demanded of Zomato in India, isn’t a magic bullet and doesn’t really work. This is well studied in academic research and basic economics: – Increased wages increases the price of the service for the end customer. This means
64
105
912
There seems to be one MLX 8-bit and one GGUF too but the results haven’t been independently verified yet afaik. May just be a benchmaxxed model, but we’ll see. https://t.co/7rctUqejhh
https://t.co/cua6PsN0ye
huggingface.co
3
0
34
Model Page: https://t.co/0hFnD0azf5 HuggingFace: https://t.co/9OGpb0asaT Paper: https://t.co/UTU9R0SINt Jian Yang scholar:
scholar.google.com
Beihang University & M-A-P; Associate Professor - Cited by 16,429 - AI2SE - Code Large Language Model - Multilingual Pre-training&Generation
1
1
35
2026: A new AI lab out of China claims 81.4% on SWE-Bench-V and 54.2% on BigCodeBench with a 40B coding LLM. IQuest Labs, like DeepSeek, comes from a quant fund Ubiquant in China with $8B AUM. Jian Yang, the leading contributor, seems to have been on the Qwen2/3 team as well!
52
57
688
Insane experience with people with zero civic sense when traveling in Vietnam > be me > on Ha Long Bay cruise in Vietnam > only one public lounge on the whole boat > 10 families quietly enjoying the scenery > six aunties in their 50s roll up > full “Bollywood wife” energy >
331
274
3K