Chris Hart

Win & help win. Co-Founder (Techstars Chicago), @gBeatapp & @V1sports | Startup Advisor 🏰

[Chris Hart @cghart]

A seminal truth so many are blind to understand.

[Naval @naval]

Using creativity to solve problems cannot be taught, and it’s ultimately what free markets pay for.

[Chris Hart @cghart]

@ClaudeDevs This is not going to age well.

[Chris Hart @cghart]

@antirez https://t.co/kZT1v6huWj

[Chris Hart @cghart]

@bcherny It's been 2-weeks, will this get fixed soon? https://t.co/82miudydsZ

[Chris Hart @cghart]

@antirez https://t.co/AsrwNySDWM

[Chris Hart @cghart]

Two contributions to @antirez's ds4 (DeepSeek V4 Flash) on NVIDIA GB10 / DGX Spark today:

• Fixed silent MTP α=0 bug on CUDA (fd-cache → NaN drafts)
• +1.5 tok/s (13.66 → 15.16) via F16 GEMV uint4 vec + q8_0 CTA-parallelism + output_a/b fuse

https://t.co/DqD1qx6PoV

[Chris Hart @cghart]

Two contributions to @antirez's ds4 (DeepSeek V4 Flash) on NVIDIA GB10 / DGX Spark today:

• Fixed silent MTP α=0 bug on CUDA (fd-cache → NaN drafts)
• +1.5 tok/s (13.66 → 15.16) via F16 GEMV uint4 vec + q8_0 CTA-parallelism + output_a/b fuse

https://t.co/DqD1qx6PoV

[Chris Hart @cghart]

@mcuban https://t.co/jekb8ewGsN

[Chris Hart @cghart]

@iotcoi @SpaceTimeViking Thank you, posting soon?

[Chris Hart @cghart]

@Tnimbus @TheAhmadOsman Good news: bjk110 v020-ngc2603 already ships with upstream TurboQuant. Ran --kv-cache-dtype turboquant_k8v4 on Qwen3.6-35B-A3B-NVFP4 — works, ~23% slower decode for 4× KV capacity. Could your agent help find when that tradeoff is net-positive for a given workload?

[Chris Hart @cghart]

Love the direction, but “simple now” still feels ~6 months out on bleeding-edge stacks. Migrated a GB10 from Avarok vLLM → bjk110 for Qwen 3.6 NVFP4 today. ~5 undocumented bug workarounds, 3 false-start tunes, judgment calls no agent makes blind.
Agents + Humans still debug.

[Ahmad @TheAhmadOsman]

Let me make local AI easy for you

Inference setup & optimization are simple now
across pretty much any Inference Engine

Just have an agent run evals overnight
sweep a few quants per GPU / inference stack
and generate bash scripts with best configs

Wake up to tuned performance

[Chris Hart @cghart]

@karpathy @WoodwayTreadmil might want to be following this..

[Chris Hart @cghart]

@sweatystartup *ApoB of 110 is not good
*ApoB < 60 should be the goal with known plaque
*Your friend is lucky — not unhealthy
*You did him a massive favor

[Chris Hart @cghart]

@arvidkahl $2 ,799 -- https://t.co/bVEWYb8vYu

[Chris Hart @cghart]

@RayFernando1337 Apple needs to double down on additional sensors/sensor fusion coupled with AI to elevate health & activity insights. IMO, innovation has stalled.

[vaajaa✨ @kerduscoksu]

he tried to clean it

[Chris Hart @cghart]

@KrissBergTweets Any before and after blood work? Specifically, A1C

[Chris Hart @cghart]

@ShaneMac Sent, thank you!

[Chris Hart @cghart]

@ShaneMac Hi, installed. Heads up on iOS, tapping on Privacy & Terms on the QR camera screen takes the user to — https://t.co/bW76bmW5SH which is displaying, “Not Found” for me.

[Chris Hart @cghart]

@JohnLegere @TMobile is in need of some John Legere love & guidance as it has become just like the other mobile Telco’s tacking on so many fees post published price.

[Chris Hart @cghart]

@mediaking It’s not just noise—it’s reality distortion. I’ve seen founders raise hundreds of millions, only to get wiped out by late-stage NY funds. One ‘TechCrunch success’ ended with the founders walking away with almost nothing.

[Chris Hart @cghart]

@MartinGTobias 😎 - does their platform enable you to review a summary or a transcript of the conversation based off of certain criteria that the AI flags (for deal flow)?