2x AI company exits to @ConstantContact & @Gladly | ex-pro IRONMAN, 6x ag podiums and Kona 3x
Los Angeles, CA
Joined June 2009
- Tweets3.9K
- Following865
- Followers685
- Likes4.5K
AI solved a century year old math problem in 88 hours
To tackle the Navier-Stokes equation models processed 150B tokens (~125B words). Reading that output would take a human 14 years non-stop, or 60 years on a 40hr work week = entire career to solve one Millennium Math problem
The millenium math proof shows AI can accomplish the equivalent of an entire lifetime career of a human, in just 4 days
The run used ~150 billion tokens, ~125 billion words, that would take a human reading 24/7 - 14 years to complete
On 8 hrs/day 5 day/wk, it becomes 60 years
Mandating every employees uses $2000/month on tokens is ridiculous. It’s a terrible metric. It encourages burn.
Try these instead:
> Automate 10 minutes of work for your role with AI every week
> Start every task with AI first, create a skill when you get the desired result
James Hotson retweeted
OMG these numbers
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision.
The longer and more complex the task, the larger Fable 5’s lead over our other models.
ALT Benchmark table titled Mythos 5 & Fable 5, comparing Claude Mythos 5 and Fable 5 against Claude Mythos Preview, Claude Opus 4.8, GPT 5.5, and Gemini 3.1 Pro.
Glean’s top line crosses $300M as AI budget-cutting becomes its major selling point techcrunch.com/2026/05/28/gl…
Dunno how @adcock_brett runs 4 companies... but its impressive
Today we’re excited to announce over $700 million in Series A funding to accelerate our ambitions to realize a generational leap in computing.
Hark is an AI lab building the most advanced, personal intelligence in the world. We're pairing our own foundation models with bespoke hardware to create a universal interface between humans and machines.
Toward this goal, we’ve raised our Series A at a $6 billion post-money valuation. In a round led by Parkway Venture Capital, demand from investors was significant and oversubscribed, with participation from NVIDIA, Align Ventures, AMD Ventures, ARK Invest, Brookfield, Greycroft, Intel Capital, Prime Movers Lab, Qualcomm Ventures, Salesforce Ventures, and Tamarack Global.
With these resources at Hark’s disposal, we’re hiring some of the best talent from across the industry. We’ve also secured the infrastructure we need to bring many of our visions to life. We’ll train our next generation of models at our new NVIDIA B200 data center.
We’re ambitious, but we recognize the incredible challenges ahead. We plan to amplify human capacity, autonomy, and joy with an adept, personalized AI assistant. This is a time when the right way is the hard way. And the right approach requires a blend of bespoke-crafted software and hardware.
This round cements our ambitions to build the world-defining AI experiences of tomorrow. And as it turns out, tomorrow isn’t so far away.
My workflow lately:
✍️ write a prompt
🔨 while LLM is thinking/working, draft a new prompt + check back every 5 min to make sure its not stuck waiting for me
🫰good result, create/update prompts into a skill
now a cascade of skills and crons run tasks daily for me
An single individual person is using $1.3M of tokens in 30 days... mind bending!
(Yes its @steipete... but its still half a trillion tokens in a month!)
Been following along the Musk suit for $150B against OpenAI
The case was dismissed today on a technicality… 3 year statue of limitations
James Hotson retweeted
When GPT-5 was released, some folks claimed AI progress was hitting a wall, whereas others said progress would continue.
GPT-5.2 was released 2 months ago. GPT-5.3-Codex was released 2 days ago and is twice as token efficient for coding. It's clear who turned out to be correct.
Launched a website in its first week its getting 200 visits per day from Google
All SEO is done with AI!!!!!!
I’ll share the super workflow with anyone who likes, comments, retweets, and follows
I’ll even help implement with first 10 people to engage
James Hotson retweeted
My AI agent almost gave me a heart attack yesterday. 💀
It autonomously drafted 21 emails to high-profile investors without asking for final permission.
The emails were surprisingly good. The risk was not.
I realized that to run in "Founder Mode" with AI, you can't just give them permission; you have to give them a conscience.
The Solution: @TaskletAI Architect
1. I built a custom Skill for Claude Code that acts as a "Safety Compiler" for autonomous agents.
2. I give a natural language goal: "Build an agent to find leads and contact them."
3. The Architect scans the intent for risks: Spending Money or External Comms.
4. It automatically injects a "Safety Constitution" into the agent's system instructions.
It hard-codes "Permission Gates" for any high-stakes action.
Now, if my agent tries to spend $1 or email a stranger, it is code-locked into a "Draft & Wait" protocol.
I’m open-sourcing the Constitution and the Architect Skill today. If you are building autonomous agents, steal this.
gist.github.com/seabass011/e…
(S/o @startupandrew and @jonnydimond for building extremely good AI agents)
