Victor Choi (@idreamer) header
Victor Choi (@idreamer) profile picture

Neither fish nor fowl

New York
Joined May 2008
Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)Photo by Victor Choi (@idreamer)
I'm a manager, and I'm writing code again. AI is why. I built a local AI harness and shared it in our internal repo. Several teams have asked to build their own on the same design. What I learned: a harness has to evolve with the codebase. The catch: I used up this month's AI token budget today 😅 How does your team use AI: prompting by hand, or building harnesses?
1
28
A leaf blows past and most cameras say "Motion Detected." Apple's new camera summaries say "A puppy is chewing on a toy." How they keep it cheap and private: a home hub filters locally, and only relevant clips go to Private Cloud Compute. woongshikchoi.substack.com/p…
18
The geo-fenced, invite-only Cybercab rollout is a smart way to scale autonomy carefully while seeding the market. As a Model Y Juniper owner I wouldn’t switch, but if I didn’t already have a car I’d strongly consider it—especially since Elon is clearly designing for a future where owners can put these vehicles on the robotaxi network and turn them into income-generating assets.
Alright chat I’m in. Here’s the deal Cybercab order page is live. Invite only. Must accept delivery to one of these cities in Texas. Only operates in those existing geofenced areas. Requires a $300 monthly subscription to operate. One vehicle per customer. “Estimated delivery” December
20
I created a channel for Claude Duet, but I cannot add my project to it, so the channel lacks my context.
1
35
Full breakdown, with benchmarks, API pricing and what builders should pick: woongshikchoi.substack.com/p… #VoiceAI #GrokVoice #VoiceAgents @ArtificialAnlys
12
Kimi 3.1 looks great. it’s catching up
1
28
Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published @Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index. Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3. Key results: ➤ Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%. ➤ Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%. ➤ Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower. Additional model details: ➤ Context window: 1M tokens. ➤ Size: 2.8T total, 104B active parameters ➤ Input modalities: Text and image ➤ Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch ➤ Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates. Congratulations to the team at @Kimi_Moonshot on the release!
220
485
341
5,475
672,050
If you look at the iPhone Duo as a cross between an iPhone and an iPad Mini, it makes a lot of sense in terms of versatility. I have to admit, I'm pretty tempted to buy one. #iphoneduo
42
OpenAI is falling behind and Anthropic is pulling away. Is GPT-6 their only hope left?
35
The secret to modern voice AI isn't one omni-model. It's a two-tier decoupled stack: A lightweight, full-duplex speech layer (GPT-Live) handles continuous WebRTC audio & interruptions, while a frontier engine (Astra) reasons asynchronously behind an RPC boundary. Zero dead air.
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.
1
91
I've been getting great use out of the Robinhood Gold Card’s 3% cash back, but now I keep getting prompted to upgrade to Platinum. Looks like the 3% cash back was just a carrot to drive user acquisition, and they’ve realized it’s a losing game for them in the long run.
80
That's a fascinating update for visionOS! Siri’s ability to "see" and interact with what’s on the virtual display is a huge leap in spatial computing. How do you think this will change the way we use AI for daily tasks?
26
Starlink isn't reliable if I connect more devices. I experience intermittent disconnection if I connect multiple devices. #starlink
22
Model Y Trunk Fridge 🧊 Bought a fridge last Black Friday for camping—super useful for groceries too! Tesla just released an official one for ~$600. Mine was $250 cheaper, but theirs supports dual-zone cooling. Curious how the official one stays powered when parked?
64
AI frontier labs don't want to be your utility provider—they want to eat the entire enterprise application layer. By building out massive Forward Deployed Engineering (FDE) teams, OpenAI and Anthropic are coming directly for the data ontology layer. youtube.com/watch?v=0A3sGymV…
1
34
The companies that thought their unglamorous data plumbing was a permanent moat are about to face an existential stack war.
11
Critical thinking is becoming the most important skill a human can have, not only because of AI but also because of the flood of unverified news these days.
16
Just canceled my GitHub Copilot annual subscription and requested a refund. The price jump was straight-up unreasonable. Now hunting for the best alternatives for a personal LLM setup — what are you actually using right now? Drop your top recommendations 👇
1
42