Jack Smies

Why this week is the best argument yet for Open models and decentralised AI An NYU mathematician spent months pushing toward a Millennium Prize problem. He did the work inside OpenAI’s Codex. Then OpenAI published a proof down the same uncommon path, after pouring roughly $22.5 million of compute into a sprint that started days after word of his progress reached them. OpenAI says it did not use his prompts. It also says it cannot rule out that de-identified product usage helped improve the models. That is the whole problem in one sentence. The prize is $1 million. The compute bill was many times that. This was not about the bounty. It was about owning the headline: our AI solved one of the hardest problems in mathematics. A customer put unpublished drafts into a closed product. The lab that owns that product has the weights, the logs policy, the compute, and the press machine. The researcher does not. You do not need a conspiracy to see the incentive. Closed frontier tools create a structural conflict: - Your best unpublished ideas live on someone else’s servers. - Those servers can train the next model. - That lab can then outspend you by two orders of magnitude. - Credit, timing, and narrative still sit with the lab. If you are a researcher, you just learned a simple rule: do not keep the work that would make your career inside a product you cannot inspect, cannot export, and cannot audit. Open models do not magically make people honest. They change the power balance. Weights can be run locally. Training data can be disclosed. Independent labs can reproduce the result without asking permission. A mathematician can use the tool without wondering whether the tool’s owner is also the competitor. Closed models gave us a breathtaking math result this week. They also gave us a reason not to trust them with the next one. Open weights are not a vibe. They are how science stays science when the instrument is also a company. #DecentralisedAI #OpenModels #OpenSource

Join me at @ai everything abu dhabi on 6-7 October 2026, where the world convenes in new intelligence.

It pains me to see what Europe is doing and regulating all innovation away but this is why you need to invest in the USA.

This is pretty bad. Any tool can be used in a good way or bad way. Limiting the use of tools usually only benefits the bad guys.

Selling something at $100. Buying it back at $86 Pretty smart. I wish I had buyers and sellers for my products like that.

Sad to see the USA shooting themselves in the foot.

Every leader needs to know some basic finance. Now that I am involved in a few audits with cxostudio.ai I am surprised to learn how little financial knowledge some leaders have. Warren Buffett once said that accounting is the language of business. You don't need to become an accountant, but you do need to be able to read the score. A leader who can't read financial statements is operating with the dashboard covered. Please check yourself. Do you know? 1) Cash flow: Cash in bank, Monthly burn rate, Runway, Operating cash flow and your Working capital requirements. 2) Income Statement: Do you understand the the relationship between Revenue, Cost of goods sold (COGS), Gross profit, Operating expenses, EBITDA, Net profit, Your deferred revenue and revenue recognition as well as amortization methods 3) Unit Economics: Customer Acquisition Cost (CAC), Lifetime Value (LTV), Payback period, Contribution margin. 4) Balance Sheet: Cash, Accounts receivable, Debt, Inventory, Accounts payable, Equity. The ratios between these and what good looks like. 5) Cap Table and Dilution: this one especially for founders: Equity ownership, Option pools, Convertible notes, SAFE notes, Preferred shares, Liquidation preferences. You can build a great company and still lose more control than you expected if you do not understand your own cap table. Raising money is not only about valuation. It is also about structure. 6) Know your metrics and best practices as here AI will have the biggest impact. ARR/MRR, Net Revenue Retention, Churn, CAC payback, GMV, Take rate, Contribution margin, Inventory turns, Gross margin, Cash conversion cycle, Utilization, ROA, EBITDA margin. If you dont understand one of these terms type them in chatgpt and learn about them. You will then see the power in the dashboards that cxostudio.ai builds and it will turbo charge your business. CFO's please add any concepts that I have forgotten.

I’m shorting monitor companies. Long AI agents. 😍 I went through this exactly. Had to buy a KVM switch to easily switch between my agents. People with 3 monitors on their desk are the dinosaurs.

The US government just told Anthropic: make your AI jailbreak-proof or you can't export it. Security experts say that's impossible. Anthropic's CEO agrees he shouldn't be the one making that call. Meanwhile open-weights models just hit #1 in frontend coding. The center of gravity is shifting fast.

Zero University - Zero Cost zero.university?ref=2HWEHN

This will be the end of many local currencies in emerging market nations plagued by high inflation, rapid devaluation, and heavy remittance reliance. By embedding a digital dollar directly into an app used by 60 million active global customers, MoneyGram accelerates "digital dollarization." This allows citizens to completely bypass weak domestic fiat currencies. Most At Risk in my view: Argentine Peso (ARS), Nigerian Naira (NGN), Turkish Lira (TRY) Venezuelan Bolívar (VES), Pakistani Rupee (PKR) If you hold significant wealth in these currencies get out now.

Similar case as with Strategy, the flywheel could stop. Or it doesnt. The market can remain irrational longer than you can remain solvent.

Recruiters, VCs, and anyone actually evaluating you are getting less and less signal from LinkedIn and Twitter these days. Between the AI slop and the endless stream of “Congratulations on your new role!” posts, it’s harder than ever to get a real sense of how someone thinks. What people are increasingly looking for is insight into your actual point of view, the problems you’re obsessed with, how you reason through them, and what you believe. Therefore post on QA sites, your own personal website or places like til.iopcap.com. With Ai is becomes easier to get a view from these anyway

Another open-weight model to try out. This one for images

We have just started building a "Company Brain" for a consultancy company. As you can imagine handling memory is an important part of the architecture and I am deep in the weeds of 'semantic', 'episodic', 'profile', 'summary' and 'tool' memory types. 😊 😅 We are using postgress vector, langGraph and LangChain but I just came across Mastra. Anyone experience with Mastra? Would appreciate a chat.

The good thing is Open Weight models are only 4 months behind. (Although Open Weight is not the same as Open Source)

Indeed who will guard the guardians. The best protection that we can have against the mis-use of AI is opensource and decentralized AI

They are getting better everyday ...