Nick

Co-Founder @Protocol_Wealth | Views are my own. Not investment or tax advice.

[JT Rose @jt_rose]

The best argument for open models is access.

New York tried to ban LLMs for legal advice. People used open source models anyway. And the advice they got, imperfect as it is, beats what most of them could afford before. Lawyers were never in most people's budget. The ban priced them back out of advice entirely.

@ccatalini called this a massive equalizer when he came on @postagixyz to talk with @sreeramkannan and the open weights letter this week makes the same bet.

Once intelligence is this cheap to distribute, restricting it just decides who gets to be smart.

[Jensen Huang @JensenHuang]

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.

https://t.co/AUKzoQ5Ikb

[Ethan Mollick @emollick]

Unexpected finding from a big study on what ChatGPT did to colleges: "once the COVID-19 disruption is modeled separately, the introduction of ChatGPT had no detectable effect on grades.. course evaluations for subject understanding, interest, and relative workload show no change"

[Haseeb ＞|＜ @hosseeb]

The dominant narrative in AI is that superhuman AI hackers are going to be a security nightmare.

I now believe the opposite: everything gets more secure after superhuman LLM hackers, because the commoditization of intelligence favors defenders over attackers.

Here's why:

• Before LLMs, defense meant covering every idea that any attacker might come up with. Human minds are diverse and attackers are uncorrelated with each other, so attack ideas had high entropy. N attackers meant N independent darts thrown in idea-space. You could outspend all attackers combined, but if even one gets lucky with and thinks of something you didn't, it's gg.

• AI collapses that entropy. Attackers and defenders alike now use the same models, which means their ideas are correlated. Even if they're using different models, models are more correlated with each other than humans are. 20 uncoordinated attackers are no longer 20 independent minds: they're running the same search 20 times, burning redundant compute, trying the same attack ideas.

• So the game reduces to spend. If you assume entropy is literally 0, the defender only needs to spend as much as the best-funded attacker, not the sum of all attackers. IOTW, attackers used to sum, now they max.

• Example: say you have North Korea, Russia, and China all trying to hack you with LLMs, and they're each spending ~$100K of compute trying to attack you ($300K total), you only need to spend ~$200K to ensure nobody gets through. This is because they're not coordinating. In the previous model, they didn't need to coordinate because the entropy in their different attack ideas were large, and thus you'd need to match their collective budgets to stay secure.

This model is obviously a simplification, and it gets compounded by other techniques that favor defenders, like formal verification and better testing.

But you can already see this playing out in crypto. Big, well-defended protocols have become more secure after superhuman AI has emerged (as I demonstrate in the QT). I believe this is the reason why. It's the same reason why AI slop in writing is so easy to spot, and I think it means the long-term cybersecurity equilibrium is actually a happy one.

[Rohan Paul @rohanpaul_ai]

"AI doesn't work on our codebase."

I hear this weekly. Same problem: they skipped context acquisition.

They pointed an agent at a repo, handed it a ticket, and got code that compiles but misunderstands the architecture.

One team built the context layer first. Two engineers rebuilt a FedEx supplier's delivery platform in 3.5 months when the original estimate was eight.

Week one, they wrote zero code. They scanned the entire repo into a knowledge graph covering modules, dependencies, data flows, and domain terms. Agents read that map before touching anything.

Every ticket ran this loop:
1. Define scope and constraints
2. Write a technical spec
3. Plan the implementation
4. Agent implements, engineer reviews
5. Test against the spec
6. Update the knowledge graph

Nothing merged unless the engineer could verify it, explain it, and debug it without the agent.

The team merged 122 PRs in 90 days with AI generating 90% of the code at $200/month in compute.

Map your repo into a knowledge graph and run this loop before you hand the agent open-ended tasks.

I recommend you to follow @mardehaym.

He runs @LimestoneHQ and posts production AI breakdowns like this.

[Ejaaz @cryptopunk7213]

if you’re wondering why intel crushed earnings (again):

- they’re fast becoming the U.S. version of tsmc. onshored in america. US gov owns 10% of it.

- intel CPUs are key to unlocking agentic ai. Agents need the ability to call tools and orchestrate actions. cpus do that.

- intel has major deals with nvidia, tesla and the largest share of onshore chip fabs in the U.S.

- since taking over, Li Bu Tan is running this company like a startup.

intels becoming a very core part of americas ai apparatus and the administration are the ones making that so 🤷🏽‍♂️

[Ethan Mollick @emollick]

I wrote the latest of my occasional guides to which AI to use right now for non-experts who want to get stuff done.

The agentic systems available to everyone are getting extremely powerful (even as the names and features continue to be really confusing): https://t.co/0H1p6QNrTd

[Andrew Ng @AndrewYNg]

Announcing OpenWorker! An open-source agent that doesn't just chat with you, but delivers finished work -- like hand you a polished document, send a slack message, or update a calendar entry.

Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert. It works across your files and everyday tools, produces the deliverable, and checks in before doing anything consequential.

OpenWorker runs on your Mac, with Windows support coming soon. It does not lock you into any one model. Bring your own API key and run it with GPT 5.6 Sol, Claude Fable, Gemini 3.6, an open weight model (like Kimi, GLM, DeepSeek, Inkling), or Ollama to keep your data local. Your data does not leave your machine except through an LLM provider and integrations that you choose.

@rohitcprasad and I are building OpenWorker because AI coworkers are an important way to get work done, and we want there to be an open, privacy-preserving, model-independent option. Check it out and let us know what you think!

Try it out: https://t.co/P0mGnI1o31 (requires your own API key)
Source code: https://t.co/NYCiTD6hSq

[Christian Catalini @ccatalini]

Timeless Silicon Valley advice, repurposed for our nation’s AI policy: in the race to AGI, obsessing over competitors is a recipe for failure.

Build an industry obsessed with solving customers’ problems. Win with the cards you’ve been dealt.

[Omid Malekan @malekanoms]

As many of you know, I've often been a critic of the various proposed laws and regs coming out of DC about crypto for years. Bad laws, I have always thought, are worse than no laws.

But I'm now calling for Congress to pass Clarity, because to this critic, the overall body of work is "pretty good."

Digital commodities are protected from securities laws written for very different assets, developers and DeFi are protected from unfair prosecution, the stablecoin yield provision leaves room for innovation, and our legal system has the tools it needs to prosecute the bad behavior the crypto industry has had too much of.

Most of all, America's place at the forefront of crypto innovation and adoption is secured.

There is room for improvement, but no bill is perfect, and this being an emerging domain, it would be foolish to let perfect be the enemy of good. We don't know enough (yet!) about this technology to create perfect rules.

To the crypto industry: We know what it's like to operate under the chaos of no laws. It usually leads to unfair outcomes and abuse. It's time to come together and support Clarity.

To the leaders of TradFi: You now know that stablecoins, tokenization, and other applications of this tech are going to be a big part of your future. But you can't succeed or even compete without rules. It's time to speak up and support Clarity.

To those worried about ethics issues: America is going to have to rethink a lot of things when it comes to conflicts of interest for elected officials. Crypto is the least of it; equities, derivatives, and prediction markets must be addressed too. Clarity is at least a step in the right direction.

To all the remaining holdouts: The alternative to passing Clarity isn't a different bill—it's no bill, at least for the foreseeable future. We are too far along the adoption curve to take that chance.

Other countries would love to see us fumble this. They've viewed America's capital markets with envy for years and view crypto as an opportunity to take market share. Let's not let them.

It's time to pass Clarity and move blockchain adoption forward.

[charlie marketplace @charliemktplace]

Uniswap Alpha leak 🧵 - The most in depth divergent loss analysis *on real data* ever done.

- Every closed position from inception to 2022-09-20
- Block level market price for ETH & BTC
- Calculated PnL, HODL value, Strategy value of every closed position
- Found ELITE LPers.

[Rohan Paul @rohanpaul_ai]

The cable routing alone deserves a moment of silence.

A view of @nvidia's Vera Rubin NVL72 cluster.

72 GPUs sit in that frame and behave as one.

Every cable in that rack sits on top of the whole planet's industrial base.

EUV machines from the Netherlands, wafers from Taiwan, memory stacks from Korea.

They etch atoms with light and ship them by the rack.

One box holding the output of four continents.

[Chubby♨️ @kimmonismus]

The competition is intensifying! The first local agent claims to be beating Hermes.

• Runs Qwen, Gemma, and Llama via llama.cpp
• Stable-prefix caching keeps long sessions cheap
• TurboQuant shrinks the KV cache by 6.4×
• Solved 37 tasks vs. Hermes’ 31

I'm a huge fan of open source. Therefore, I'm delighted to see significantly more competition here. Local and open source is the future.

[Ethan Mollick @emollick]

Cool paper looking at how AIs solve unbounded, complex business problems in many fields by testing how well they can crack the cases we use to teach MBAs in business school:
1) AI already does extremely well across diverse business topics
2) Models are improving rapidly with time

[Naval @naval]

When software was expensive - thin, horizontal, best-of-breed software stacks extracted rents across every business.

Now that software is cheap - value moves to vertically integrated businesses that deliver opinionated end-to-end experiences.

[Chris Dixon @cdixon]

https://t.co/twoyc1x91x

[OpenZeppelin @OpenZeppelin]

For banks, going onchain is no longer the hard part. Making it defensible is.

Tokenized deposits, settlement, and custody now have to meet the same security bar as everything else a bank runs.

See how OpenZeppelin secures onchain programs end to end ↓
https://t.co/itXexBPflH

[Omid Malekan @malekanoms]

The missing bit: instant settlement can’t work if exchanges like the LSE (or any other) try to keep the intermediated structure. The cost of capital is too high.

Yet another reason vapid tokenization by TradFi clearinghouses involving the same brokers as today is bound to fail.

[The Kobeissi Letter @KobeissiLetter]

BREAKING: AMD, $AMD, and Anthropic have signed a deal for "tens of billions of Dollars" worth of AI servers, per WSJ.

Details include:

1. Anthropic will purchase up to 2 gigawatts of AMD’s latest-generation chips, the Instinct MI450, starting in the first half of 2027

2. AMD will invest up to $5 billion in Anthropic as certain deployment milestones are met

3. AMD is also in talks to provide a financial backstop for Anthropic’s future data-center leases as well

AMD continues to look for ways to strengthen its positioning against Nvidia and others.

[Ophir Gottlieb @OphirGottlieb]

$CBRS only the beginning IMHO

CrowdStrike and Cerebras Partner to Power AI Detection and Response on the World's Fastest Inference

"CrowdStrike will leverage Cerebras’s industry-leading inference speed to help power Falcon AI Detection and Response (AIDR), while Cerebras standardizes on the CrowdStrike Falcon® platform to secure its business. Together, the companies are pairing the inference speed with AI-native security for enterprises building and deploying AI at scale."