Andrei Oros

Why is the creator of OpenCode pretty skeptical about AI productivity gains, and the hype around AI? A very conversation @thdxr (and lots of truth bombs:) Timestamps: 00:00 Intro 07:03 Dax’s path into tech 09:04 Early startup experience 13:16 Getting involved with open source 16:13 OpenCode 23:17 Anthropic banning OpenCode 30:34 From terminal to GUI 32:34 OpenCode’s business model 36:33 Why inference is profitable 39:11 GPU bottlenecks 40:54 AI hype 45:50 AI spending 48:47 Dax’s memo 55:41 Dax’s skepticism of predictions 58:58 Engineering culture at OpenCode 1:02:38 How building works at OpenCode 1:05:36 Taste and quality 1:11:32 Dax’s work setup 1:12:35 The role of engineers and EMs 1:15:50 Advice for engineers 1:18:12 Book recommendation Brought to you by: • @AntithesisHQ – verify your system’s correctness without human review or traditional integration tests – and avoid bugs or outages antithesis.com/pragmatic • @WorkOS – everything you need to make your app enterprise ready workos.com • @turbopuffer – a vector and full-text search engine built on object storage. It’s fast, cheap, and extremely scalable turbopuffer.com/pragmatic Three interesting thoughts from Dax: 1. No AI-native coding agent company is “winning” by being better with AI. Dax says that none of OpenCode’s competitors are crushing them, and that nobody is using AI so well that others cannot compete. 2. Most software engineers profit from AI as time gained, not increased output — unless you change incentives! Dax says the natural way for software engineers to “cash out” their AI tooling gains is with time savings, by doing the same work as before, but faster. Until compensation and motivation structures change, most teams should expect output to stay flat while engineers go home earlier. There’s nothing wrong with this, but AI vendors sell a different outcome to CFOs: increased output. 3. AI code generation mutes the “guilt” of doing the wrong thing, but this builds up tech debt. Pre-AI, writing a hack felt bad, the second time it felt really bad, and by the third time you’d often just refactor in order to fix up the code. Now, the agent hides the hack, which skews devs’ judgment and results in less tech debt being cleaned up.

@bcherny did my part to help students with this free course andreioros.com/blog/agentic-a…

Big fan of teaching more people the basics of using Claude Code in an accessible way. So much of the world has not yet used agents. There's a lot of opportunity to level the playing field and expand access.

Scientists say they have found the strongest evidence yet linking a common virus directly to skin cancer. Researchers studying a 34-year-old woman with repeated cases of skin cancer discovered that a form of human papillomavirus called beta-HPV had inserted itself into the DNA of her tumor cells. The finding surprised scientists because this type of HPV was previously believed to only increase cancer risk indirectly by making skin more vulnerable to damage from ultraviolet light. Instead, the study suggests the virus may actively help certain cancers grow and survive. The woman had an aggressive form of cutaneous squamous cell carcinoma, one of the most common types of skin cancer. Despite surgeries and immunotherapy treatments, the tumors kept returning. When scientists carried out a detailed genetic analysis, they found the virus embedded inside the cancer cells themselves. It was also producing proteins that appeared to support the tumor and help it continue growing. Researchers said this is the first time beta-HPV has been seen integrating into human DNA in a way that may directly sustain a cancer. The patient also had a rare inherited immune disorder that weakened her T cells, which normally help the body fight HPV infections. Scientists believe this weakened immune system allowed the virus to spread more deeply into skin cells and contribute to the cancer’s development. Doctors later treated her with a bone marrow stem cell transplant to rebuild her immune system using healthy donor cells. After the procedure, her skin cancer disappeared along with several other HPV-related conditions. During the next three years, none of the diseases returned. Researchers stress that sun exposure and ultraviolet radiation are still the leading causes of this type of skin cancer. However, the findings suggest viruses may play a much larger role in some cancers than previously understood, especially in people with weakened immune systems.

Yann LeCun says that within a year to 18 months, we'll have a general method for training hierarchical world models These models would learn from video and real-world data, then help plan actions in robotics, healthcare, and other areas "then scale them toward a universal world model"

Alzheimer’s may be linked to gum bacteria, new research shows. Scientists have repeatedly found Porphyromonas gingivalis—the chief bacterium that causes periodontitis—inside the brains of people who died with Alzheimer’s. When researchers deliberately infected mice with this oral bacterium, the animals rapidly developed key Alzheimer’s pathology, including the buildup of amyloid-beta plaques. Perhaps most alarming, the bacteria’s toxic enzymes have been detected in the brains of people showing early Alzheimer’s changes years before memory loss or other symptoms appear, suggesting the infection may quietly initiate damage long in advance. These discoveries have sparked serious interest in new treatment approaches. An experimental drug called COR388 (from the company Cortexyme) has already succeeded in lowering both bacterial load and amyloid-beta levels in preclinical models. Although large human trials are still needed, the evidence is mounting that at least some cases of Alzheimer’s may have an infectious trigger rather than being purely degenerative. [Dominy, S. S., et al. "Porphyromonas gingivalis in Alzheimer’s disease brains: Evidence for disease causation and treatment with small-molecule inhibitors", Science Advances, 5(1), eaau3333]

no #skininthegame

A visualization showing how gravity curves spacetime [🎞️ Daniel Ibarra]

It’s extremely hard to “do a little disclosure” and avoid the tight scientific/societal questions. The only way this has worked has been starving smart technical people for any detail at all. What I will be watching: Everyone: “Did the government psy-op us all with BS? If so: Why????” Biologists: “You said the government has alien biologics? Do they have eukaryotic cells? If so what do we know about their histones, matrilineal mitochondrial dna, hemoglobin atp synthase??? What is their placement on the phylogenetic tree? Tetrapods??? Isolates?? If not eucaryotic, how are their cells/tissues organized? Do they use proteins?? What are their body plans??” Linguists: “What is the structure of their communication schemes? Can it be mapped onto a generalized human grammar? Or are our languages not expressive enough to cover their languages? Do they use sound or light or some other wave to transmit/receive? Do they have an analog of music??” Physicists: “How do the standard model and general relativity appear as effective theories/lagrangians of the alien understanding of the cosmic waves, media and fabric for lack of better terms? How many dimensions are there, are they engineering accessible and how many new ones are temporal? Are there new forms of energy corresponding to these new degrees of freedom? Is the speed of light gameable?” Civil Libertarians: “How many innocent lives were ruined keeping this secret? Did we fake a lot of this?? Who authorized the lying, discrediting and possible wet work?” NatSec: “Are we owned by an unknown force? Will it now be trivial easy for everyone to make WMD from new discoveries?” Etc. This is not going to stay controlled if it is at all specific. As soon as there are any specifics I guess that the game changes character instantly and goes into high gear with totally different players.

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

Persistent memory is the Achilles heel of AI. Engramme’s Large Memory Models (LMMs) empower every app with persistent memory. Google solved search. OpenAI solved language. Engramme solved memory. Join beta: engramme.com/signup

Your chances of getting robbed in Romania is around 10x lower than most Western European & US cities. The Romanian government (despite all its problems) also robs you much less. You can earn €1M / year as a self employed person and pay 10% flat tax. Earn €1M on stocks and pay 3% capital gains tax. Sell €1M in real estate and pay 1% tax.

Our team ran a verifiable quantum algorithm that probes how parts of a quantum system interact, from molecules to magnets and beyond. On our Willow chip, it ran 13,000× faster than the best classical supercomputers. A first in quantum computing → goo.gle/42z9E2d

99.9% of the world is completely unaware what is going to hit them in <5yrs: incredibly powerful, cheap intelligence. The way we are all addicted to information, glued to our screen 24x7, we are about to get addicted to hyper-intelligence. Work is growing 10x from here, unf, not reduce.

23 years old with no advanced mathematics training solves Erdős problem with ChatGPT Pro. "What’s beginning to emerge is that the problem was maybe easier than expected, and it was like there was some kind of mental block.”-Terence Tao scientificamerican.com/article/amateu…

Singapore and Japan are both high-trust societies, despite one being highly diverse and the other highly homogeneous. So trust isn’t about diversity or homogeneity. What they both share is strong individual accountability and strong norms against imposing costs on others.

AI is not scary, but this is.

Congrats to @10x_Science on their $4.8M seed! They're building AI for molecular-level protein characterization, a process that today requires specialized scientists spending weeks or months manually interpreting complex data. Their platform delivers the same insights in minutes. techcrunch.com/2026/04/22/ai-…

DeepSeek-V4-Pro 🔹 Enhanced Agentic Capabilities: Open-source SOTA in Agentic Coding benchmarks. 🔹 Rich World Knowledge: Leads all current open models, trailing only Gemini-3.1-Pro. 🔹 World-Class Reasoning: Beats all current open models in Math/STEM/Coding, rivaling top closed-source models. 2/n

What's especially cool is that the flagellar motor exploits the indivisibility of 5 by 2 to run on a fuel of rising entropy as protons diffuse into cells. For me as a physics and math person, this is biology at its best. quantamagazine.org/what-physical-…