building @hone. exploring the matmultiverse
San Francisco
Joined May 2020
- Tweets77
- Following471
- Followers278
- Likes511
Pinned Tweet
When life gives you codex and free will
Of course you make yourself a massive LED Jumbotron in the office to monitor everyone who progresses through your puzzle
mahesh retweeted
Tina front-ran our Engines announcement by a few weeks. Fortunately she recorded a new version for us.
Frontier models can hill-climb almost any goal that is well specified and verifiable.
Most work inside a company is neither. The goals live in people's heads, and there is no way to check a change before it touches real work.
Hone's engines are built exactly for this. They learn implicit goals from users and their environment, then propose improvements to their own harness, promoting the ones that don't introduce regressions in simulation.
We wrote up how this these simulations works: hone.com/blog/closing-the-ga…
If you want to work on the problems that get more valuable with each model release, we're always looking for the brightest talent: hone.com/careers/
Today Hone launches Engines: AI that owns business outcomes.
Chatbots give answers. Agents do tasks. Engines deliver outcomes, and we hold them responsible for creating real value, measured in actual dollars instead of tokens or vibes.
You point it at a business metric, and the Engine figures out the rest. It onboards itself, builds the necessary agents, skills, and memories, collaborates with the right people, and gets better with experience.
From AI leaders to global enterprises, our customers have already used Engines to:
• Uncover millions in procurement savings
• Optimize product funnels
• Manage credit risk
• Increase qualified pipeline in new territories
• Accelerate implementations and improve account management
To scale this impact across the economy, we’ve raised a $60M seed led by Benchmark and Index with participation from Elad Gil, Hanabi, Definition, Diffusion, Lux, SV Angel, and Align.
AI’s raw capabilities are far ahead of the value it creates in the real world, and Engines will help close the gap.
Here’s how they work…
overintellectualizing the hone.com landing page reveal animation to be maximally dopaminergic
smoothstep() strikes again (h/t @iquilezles)
the models of today still seem to have very poor theory of mind..
pic is an excerpt from the claude code system prompt regarding delegation semantics, where they explicitly urge the model to "never delegate understanding", "Brief the agent like a smart colleague who just walked into the room", etc.
models tend to either vastly over- or under- delegate context to their peers/"minions" without additional guidance. unclear if this improves with larger pretaining bases that would have the higher 'raw iq' to better think from another agent's perspective, or if post-training with the right collaboration incentives would inherently develop the capability.
people go through y e a r s of school to learn how to communicate their ideas effectively (write), why don't LLMs do the same? (if we imagine a future of multi-agent collaboration)
conspicuous computation
I predict that in 5-10 years every household will have a local AI server that serves all the AI workload of the house, digital or robotics. It’ll be as common as a WiFi router or a fridge
it creates a private mesh network to link all the devices in the house and does inference completely locally, with built in data vault resistant to remote data hacking or physical tampering. That way you can have every single piece of information about you and your family stored in there and never leaves it, and AI can use them to provide most accurate and customized services
a latent space is worth a thousand* videos
a video is worth a thousand pictures
a picture is worth a thousand words
as yann lecun loves to say:
"LLM: 1E13 tokens x 0.75 word/token x 2 bytes/token = 1E13 bytes. 4 year old child: 16k wake hours x 3600 s/hour x 1E6 optical nerve fibers x 2 eyes x 10 bytes/s = 1E15 bytes. In 4 years, a child has seen 50 times more data than the biggest LLMs."
excited for robotics to force AI to learn from the firehose like humans do
Do not try the espresso shot + Diet Coke thing unless you want a vile concoction you would’ve gotten paid five bucks at the middle school lunch table to down
you're not dogfooding like i am bro. you need to be dogfood-maxxing bro. i'm raw steak fed; you're kibble-minded bro.
"what if our agents had an MCP where they can write code in a dedicated sdk to put stuff on the jumbotron"
