VP, Product at https://teapaw.org/t.co/EDV8xeP8um | Building AI Brokerage Platform
Salt Lake City, UT
Joined March 2012
- Tweets8.5K
- Following2K
- Followers1.5K
- Likes22.3K
Ryan Casey retweeted
Cheaper tokens don't mean cheaper bills
Since May:
- Average token prices are down 67%
- Token spend is up more than 120%
Charts of the Week: a16z.news/p/charts-of-the-we…
The "Rack Spread" is unfolding, token prices are falling while GPU prices climb
Change since mid-July:
- B300 GPU rentals up 106%
- B200 up 66%
- H100 up 17%
- Average token prices down 58%
Charts of the Week: a16z.news/p/charts-of-the-we…
Ryan Casey retweeted
The length of software tasks AI agents can complete has doubled roughly every 7 months since 2019, and recent models are moving even faster, from 30 seconds of human work in 2022 to a full workday today.
Yet inside most companies, AI still isn't trusted to reliably update records in Salesforce. "AI does not have an intelligence problem," says Databricks CEO Ali Ghodsi. "It has a context problem."
Ali's answer is what he calls a "system of record for agents."
From the latest Profiles of the Future by @bhorowitz and @mikemcg0 on Databricks & the Future of the Firm: a16z.news/p/databricks-and-t…
Ryan Casey retweeted
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Ryan Casey retweeted
Kalanick thought he was in survival mode when he had 20,000 people working for him. Ben Horowitz says that's one of the most dangerous ideas a founder can carry.
"He thought he was in survival mode when he had 20,000 people working for him, and that idea is dangerous when you have such a big organization."
The instinct that gets you to 20,000 employees is the same instinct that can wreck the organization at that scale. Not because the founder goes soft, but because the behavior gets amplified down eight levels of management.
The obvious lesson is that founders need to scale their leadership style. The less obvious lesson is about signal amplification.
Every behavior a founder normalizes gets replicated and distorted down the org chart. The founder pushing to the edge reads as intensity at the top. By the time it reaches a middle manager under pressure, it can read as permission. The problem was never the founder's edge. It was the absence of a system that kept the edge from becoming a norm for people who didn't have the context to use it right.
Ryan Casey retweeted
Token prices keep falling and GPU rental prices keep rising
Intelligence gets cheaper, usage goes up, and a model ends up in every piece of software
TypeSafe AI's Diogo Almeida on why Jev is called Jev, and what cheap intelligence finally changes about software:
"It's wild that AI is so cool and software has been unchanged in 10 years... The most we can do is add a little chatbot in the side sometimes that can take actions, but not all actions."
"TypeSafe is making AI for software. We want to make AI powerful, not just for humans in the loop, but to actually build real software."
"It's about the laziness to spend 10 hours to do the five-minute task instantly and to never have to do it again. It should be an ROI decision for people who automate stuff. I would just like it to be automatable."
"I think that people will just make new kinds of work, hence the Jev in Jevons. New kinds of work once that stuff is doable."
"As we have lots of intelligence at lots of different cost and speed trade-offs, the super systems-y types will be making trade-offs. Jev's gonna be 1000x too smart for them. They just want an approximate guess to optimistically route here and there."
Martin Casado: "The good news is we get to rebuild systems again. We have a new primitive. It's kind of a new way of thinking about doing software."
@CompleteSkeptic @typesafeai @bhorowitz @martin_casado
Ryan Casey retweeted
The Head of Engineering when he sees PMs pushing vibe slop to production but also has a mortgage to pay
Ryan Casey retweeted
AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023.
That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.
Ryan Casey retweeted
HEARTBREAK FOR A UTAH FAMILY: Nine children are now without their mother after Rebeca Hafen was hit by a car while out for a walk, according to a fundraiser organized for her family.
Rebeca was late in her pregnancy, and doctors performed an emergency delivery after she was rushed to the hospital. The newborn survived and is now being cared for in the NICU.
Rebeca later died from her injuries, leaving behind her husband, Tyson Hafen, and their nine children.
A fundraiser organized by Kameron is now helping the family from Moroni, Utah with medical expenses, the newborn’s needs and caring for the children as Tyson navigates hospital visits and the sudden loss of his wife.
The family is asking for prayers, support and help sharing their story during an incredibly difficult time.
Ryan Casey retweeted
The new Atlas world model is literally black magic. This shouldn't be possible from just 3 cameras 🤯
Ryan Casey retweeted
for context we built a $SPCX financial model that performed 5,000 monte carlo runs over 15 years. We constantly update the model (have another update coming in a few weeks) so its not intended be final.
But with many purposefully conservative assumptions (like embarrassingly low growth in AI apps like grok/cursor) the model showed one thing consistently. The upside here is far bigger than its downside.
This screenshot below is showing the current valuation (red circle). The red and green dots are the each one of 5,000 simulations, run at the 2035 timeline using our most conservative valuation method a Gordan DCF. the upper bound of these dots lands in the 10s of Trillions.
if you have a mindset that moves beyond quarter to quarter then this is obvious.
yes - we (as co and individuals) have spacex positions, so take this only as information (NFA) and I want to be clear that our bias is out in the open. if anyone tries to tell you they don't have a bias then you can assume they make money via clicks/views.
Ryan Casey retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
Ryan Casey retweeted
The first Vera Rubin clusters are here!
Yesterday, @IneffableLabs took delivery of their Vera Rubin NVL72 cluster from @googlecloud @nvidia
The AI frontier jumps forward by yet another generation of hardware. Acceleration continues.
Ryan Casey retweeted
A stunning piece of engineering!!
1X unveils the new humanoid hand for NEO
- 25 degrees of freedom: 22 fully actuated in the fingers and palm, plus 3 at the wrist.
- The DoF are distributed anatomically rather than evenly, deliberately biased toward a thumb that genuinely opposes the fingers.
- In-house tendon-driven, quasi-direct-drive running low gear ratios of ~5:1 to 15:1 vs the typical 100:1–200:1.
- Motors live in the forearm and pull tendons through the wrist. This keeps the hand light and its inertia low while producing high forces.
Sensing
- All 25 DoF are natively force-controlled and fully backdrivable. Every joint doubles as a force sensor.
- Very important, closed-loop proprioception: it always knows its own pose and effort without looking.
- Tactile skin across the fingertips and surfaces measuring contact and shear. This helps with adaptive gripping in real time.
Safety and durability
- IP68 waterproof and food-safe, so it can wash its own hands.
- Compliant by construction: the low gear ratios, tendon drive, and low distal inertia let external impacts safely backdrive the fingers. It yields when hit by a hammer or caught in a drawer.
- Full finger assemblies validated to millions of cycles.
Manufacturing
- Deep vertical integration: in-house motors, custom electronics, and tendon systems.
- Hundreds built already, with capacity to produce 10,000 hands this year.
"An API to the Physical World"
Ryan Casey retweeted
My third interview with @Isaiah_p_Taylor, Founder of @ValarAtomics.
Valar just became the first startup in history to power an NVIDIA Blackwell with a nuclear reactor.
0:32 Manufacturing nuclear reactors is a SpaceX style problem
2:58 Scaling from 1 reactor to 10
4:31 Why it’s so tempting to iterate on paper instead of actually building reactors
7:37 Having contact with reality - modular shielding
10:09 Steel is cheaper than software engineers
11:09 Why the idiot index is so high in nuclear
16:02 Becoming a nuclear company
18:57 It's important not to lie to yourself
21:55 Obsessing over critical path
24:54 Nuclear reactors are relatively simple
30:35 Safety is a function of iteration speed
30:55 Why Valar had to vertically integrate
35:01 Designing a Toyota Camry vs a Ferrari
38:07 Running through 1-way doors
41:13 Pulling rabbits out of a hat
46:13 Wartime mode
48:49 Being willing to do unreasonable things
51:20 How Elon injects urgency into his companies
56:56 Relentlessly chasing the goal of making energy 10x cheaper
58:32 What’s changed over the last 3yrs
1:00:51 Designing safe nuclear reactors
1:13:05 Learning from SpaceX - optimizing for scale
1:18:07 Having a high tolerance for looking dumb
1:19:50 How Palmer Luckey handles hit pieces
1:20:55 Inevitable time
1:25:26 Predicting unknown bottlenecks
1:28:08 Scaling
1:29:45 July 4th will be a rebirth for nuclear in the United States
