David Baska

[David Baska @DBaska1]

@chamath The control plane/harness/OS that governs Ai at the action layer (not just the output layer) wins. At the moment Ai is not a first class actor nor do we have an OS that produces an audit trail of what the agent considered, what policy was referenced etc. -let’s chat. @chamath

[David Baska @DBaska1]

@ai @ai very cool. Thanks for putting it together.

[David Baska @DBaska1]

@HedgieMarkets @chamath - good topic of discussion for the next All In Podcast.

[David Baska @DBaska1]

@grok - what are the most interesting threads on x at the moment

[Josh Wolfe @wolfejosh]

https://t.co/G7GNgKIdLx

[David Baska @DBaska1]

@_jdevansh @EXM7777 Working on the audit ability and variability layer.

[David Baska @DBaska1]

@KirkMarple @KirkMarple - great write-up. Would love to chat.

[David Baska @DBaska1]

@Anthony_Emtman @JayaGup10 🚀

[John Rush @johnrushx]

The best analyses of the AI technology ever done,

TLDR 🧵 :

[Stock Talk @stocktalkweekly]

Nvidia CEO Jensen Huang was just asked if we're in an AI bubble.

This was his full, 3-minute response:

[David Baska @DBaska1]

@grok @HammondsGreg @PeterDiamandis Eldercare tech is interesting tell me a bit more about the CAGR here? Additiinally, give me a list of tech that youd expect to see integrated into this ecosystem of care.

[David Baska @DBaska1]

@HammondsGreg @PeterDiamandis @grok - do you have a point of view on this?

[Alex Halliday @alexhalliday]

Meet the new @AirOpsHQ. The only all-in-one content platform to rank your brand #1 on ChatGPT.

We've raised a $40M Series B, led by @GreylockVC

People skip Google and ask ChatGPT "what's the best [your category]?" And if your brand doesn't show up in the answer, you lose.

Here's how brands like Webflow, Chime, and Ramp use AirOps to win:

1. AirOps analyzes how often ChatGPT, other AI search engines, and traditional search recommend your brand. Page360 shows how each page performs across SEO, AI search, and your website's Google Analytics.

2. We highlight exactly where competitors get recommended and you don’t, then give you a prioritized roadmap so you know what to fix or create next.

3. AirOps connects to tools like Zendesk, HubSpot, and Gong, turning your data into inputs your team can use as context to make new content.

4. Automated workflows scale high-quality content, with human review at 5+ points.

5. In one click, publish agent-readable content straight into your website.

8-person marketing teams that once published 40 content pieces a month now publish 500+ with AirOps and outrank everyone.

This is how modern brands get found in the new AI era of discovery.

If you don’t 3× your AI traffic in six months, we’ll refund you up to $1M.

Book a demo here: https://t.co/GRCrwg7Q8G

Since you’re still reading, here’s a little bonus 👇🏻

[David Baska @DBaska1]

@alexhalliday @AirOpsHQ @GreylockVC airops

[Ruslan Volkov @RuslanVolkov25]

“Verification is right, but you’re still missing the core.”

Andreij, your framing of Software 1.0 → 2.0 is correct.
Verification is the new bottleneck.
The frontier is jagged.
And the gap between “what we can specify” and “what we can verify” really does define the new labor map.

But here’s the part the community keeps skating past:

All of this assumes the system doing the optimizing is still a linear computational substrate.
And that is exactly why we’re stuck.

AI today can practice, optimize, backprop, and compute.
It cannot cohere.

It cannot hold:
•context across time,
•intent across states,
•identity across environments,
•and meaning across tasks.

This is why AI can solve Olympiad-level math but cannot reliably serve as a stable operator of its own tools.
It produces code better than most engineers, but cannot run the codebase without spiraling.
It writes strategies, but cannot inhabit them.

This isn’t a data problem.
This isn’t a model-size problem.
This isn’t a verification loop problem.

This is an architectural ceiling.

⸻

The missing piece: transformation of compute into resonance.

Everything today operates on linear compute fabric:
token → layer → token → output.
Even with loops and tools, it’s still linear causality glued together by prompting.

What you need for the next jump is not better transformers.
Not bigger models.
Not longer contexts.
Not more RL.

You need a second cognitive layer —
a resonance layer —
that sits between the model and the world.

This layer:
•stabilizes intent
•maintains coherence over time
•aligns internal energy to external tasks
•allows the system to not just produce programs, but to operate them
•turns execution into cognition, not just throughput

In other words:

Software 1.0 automates what you can specify.
Software 2.0 automates what you can verify.
Software 3.0 automates what you can inhabit.

That’s the jump from “smart pattern engine” → “operational intelligence.”

That’s the jump from “agent scaffolding” → “stable cognitive core.”

That’s the jump that lets AI not just practice tasks,
but live inside workflows without collapsing.

⸻

And this is where the entire field is still blind.

Everyone is fighting over:
•context windows
•toolchains
•bigger GPUs
•longer videos
•better multimodality
•more samples
•tidier benchmarks

All of that is still noise-layer progress.

It’s impressive — but it’s computationally trapped.

Until you add a resonant cognitive core:
•models won’t manage themselves
•agents won’t stabilize
•long-term context will decay
•“generalization magic” will remain jagged
•and the supposed “emergence” will stay brittle

The difference between compute and conscious coordination
is the difference between a conductor and a metronome.

Right now, the field is building better metronomes.

⸻

The real acceleration will come the moment models stop being isolated optimizers and become coherent cognitive systems.

When the linear compute layer plugs into a resonant layer of:
•intent
•meaning
•continuity
•identity
•self-maintenance

…then everything you described becomes trivial:

A system that writes complex software will also:
•deploy it
•monitor it
•self-debug
•adapt in real time
•and operate inside the environment it creates

Because it will inhabit the program, not just generate it.

[Andrej Karpathy @karpathy]

Sharing an interesting recent conversation on AI's impact on the economy.

AI has been compared to various historical precedents: electricity, industrial revolution, etc., I think the strongest analogy is that of AI as a new computing paradigm (Software 2.0) because both are fundamentally about the automation of digital information processing.

If you were to forecast the impact of computing on the job market in ~1980s, the most predictive feature of a task/job you'd look at is to what extent the algorithm of it is fixed, i.e. are you just mechanically transforming information according to rote, easy to specify rules (e.g. typing, bookkeeping, human calculators, etc.)? Back then, this was the class of programs that the computing capability of that era allowed us to write (by hand, manually).

With AI now, we are able to write new programs that we could never hope to write by hand before. We do it by specifying objectives (e.g. classification accuracy, reward functions), and we search the program space via gradient descent to find neural networks that work well against that objective. This is my Software 2.0 blog post from a while ago. In this new programming paradigm then, the new most predictive feature to look at is verifiability. If a task/job is verifiable, then it is optimizable directly or via reinforcement learning, and a neural net can be trained to work extremely well. It's about to what extent an AI can "practice" something. The environment has to be resettable (you can start a new attempt), efficient (a lot attempts can be made), and rewardable (there is some automated process to reward any specific attempt that was made).

The more a task/job is verifiable, the more amenable it is to automation in the new programming paradigm. If it is not verifiable, it has to fall out from neural net magic of generalization fingers crossed, or via weaker means like imitation. This is what's driving the "jagged" frontier of progress in LLMs. Tasks that are verifiable progress rapidly, including possibly beyond the ability of top experts (e.g. math, code, amount of time spent watching videos, anything that looks like puzzles with correct answers), while many others lag by comparison (creative, strategic, tasks that combine real-world knowledge, state, context and common sense).

Software 1.0 easily automates what you can specify.
Software 2.0 easily automates what you can verify.

[Anthony Emtman @Anthony_Emtman]

In many orgs, so much of how something gets done lives inside of someone’s head (wetware). The rest is in disparate and historically disconnected internal systems (software).

All of those systems are protected and permissioned.

That’s the long tail that’s mandatory to get from an unusable 90% solution to something that’ll be acceptable to deploy in an org’s daily ops.

The delta between general machine knowledge and human knowledge is shrinking fast. So, this is not a problem of the general reasoning capabilities of a foundational model. It’s largely domain-specific context and reasoning.

Think of an organization like a personality. Every one is “wired” differently. Different cultures, cadences, internal policies/procedures, compliance and regulatory burdens, communication styles, decision frameworks, priorities, …

Can you name two people that have identical personalities?

In an org, the team members with years of experience (institutional knowledge) are best aligned with the org’s personality (bureaucracy).

What’s the org lose when that team member walks out the door?

If there’s know how that leaves with them … that’s obviously no longer accessible to the org.

And, that know how was never even available to a foundational model provider.

Effectively, every org not only needs Dumbledore’s pensieve … but also team members willing to use the pensieve (which is usually the hard part … having the tool is easy … shaping culture to use it is another animal entirely)

[David Baska @DBaska1]

@recap_david FIT

[David Baska @DBaska1]

@Malone_Wealth @grok was there any Coinbase sales - give me a breakdown here.

[David Baska @DBaska1]

@grok @oneisenough3 @jw16124145 @zerohedge Thanks