The Iran War ceasefire has collapsed.
https://apnews.com/article/trump-iran-war-ceasefire-strikes-c45111ed270afa7dac285016ce07362f
‘Mr Trump finds himself back where he started three months ago: neither carrots nor sticks can persuade Iran to reopen the strait. Yet the stand-off has real costs for Iran, too. Some hawks in Tehran argue that relinquishing their grip on Hormuz would deprive them of leverage in a future conflict. That misreads reality. So long as Iran has missiles and drones, it can shut the strait at will. Control of Hormuz matters to Iran today as a means, not an end: it is only valuable in trade for tangible benefits.
With sanctions restored and the blockade resuming, Iran has already lost its immediate gains from the mou. It cannot charge tolls on vessels using the strait in peacetime unless Oman agrees to collaborate. Meanwhile, much of the crude it exported during the first weeks of the deal is sitting unsold: Chinese refiners are instead buying from Arab states, which have priced their oil at a discount to Iran’s.
Neither side has a military route to achieving its goals. Mr Trump cannot bomb the strait open; Iran cannot profit if it is shut. The least bad option for both is the deal they are busy shooting to bits.’
https://www.economist.com/middle-east-and-africa/2026/07/14/donald-trump-has-no-good-options-for-reopening-the-strait-of-hormuz
Mahmoud Ahmadinejad is back from the dead!
https://www.israelnationalnews.com/news/429766
‘The hostility directed at top officials during the funeral reflects a theory that has been gaining traction within the Islamic Republic’s most radical factions for months: that Iran’s wartime leaders who negotiated and signed the agreement with Washington are staging a soft coup against the Islamic Republic and its revolutionary ideals while the new supreme leader remains largely invisible for fear of his life – or, as some have suggested, because he is incapacitated.’
https://www.cnn.com/2026/07/18/middleeast/iran-hardliners-coup-warning-us-truce-intl-cmd
‘US military used three Corsair sea drones with explosives to strike an Iranian naval base, marking their first combat use.’
https://youtu.be/m0evF4m2510
This prediction (thankfully) didn’t happen: ‘Thus, by the middle of July, the full extent of temporary buffers will have been exhausted, with an overall market adjustment of 7.1 mb/d, roughly 16% of global crude oil trade, needing to be absorbed.’
https://www.brookings.edu/articles/the-timing-of-the-impending-crude-crisis/
Ukrainian drone and missile attacks on Russian military logistics and civilian infrastructure have been very effective.
https://www.aljazeera.com/news/2026/7/3/russian-advance-collapses-in-ukraine-as-anxiety-rises-in-moscow
‘Ukraine Claims Scores Of Russian Ships Struck In Sea Of Azov’
https://www.twz.com/news-features/ukraine-claims-scores-of-russian-ships-struck-in-sea-of-azov
‘Russia launches major ballistic missile attack on Ukrainian cities’
https://www.bbc.com/news/articles/c2el7xpnzrpo
Rosario Randazzo was “all about the job.”
https://medium.com/@factsfeed/a-brief-story-of-an-italian-war-hero-96644a10c6f6
In WWII, the Germans were bold, smart, and used their inferior resources efficiently against the Allies. One example of this was their ambushing of several U.S. bombers over England on the night of 22/23 April 1944.
https://youtu.be/V5WKz_TZaXM?si=GJLtUJcScvZcERKr
‘Despite Ordnance descriptions of the knife as being designed for hand-to-hand warfare, the M3 did not receive universal praise as a close-quarters fighting knife upon issue to combat units.’
https://en.wikipedia.org/wiki/M3_trench_knife
Note to self: When the apocalypse happens, don’t wear your WW2 Romanian M39 Steel Helmet into combat.
https://youtu.be/fTSTWwRgSuM
North Korea’s construction of destroyers is a major misallocation of resources. Even if they managed to build ten of them, they’d all be sunk on the first day of a war with the U.S. (at very little or no risk to our own forces) because we have such a huge military advantage.
https://youtu.be/Fc8eQf9YZbk
‘China Builds Full-Scale US Warship Replica For Missile Target Practice’
https://www.ndtv.com/world-news/china-builds-full-scale-us-warship-replica-for-missile-target-practice-11761338
The notion that all weapons-related stuff the Russians make is robust is false, and the “OKP-7” rifle sight is a great example.
https://youtu.be/v6vy6Ryta6c?si=bXKglHi95JnGOexC
In 1991, a Cuban fighter pilot defected to the U.S. by flying his warplane to Florida. After two years of failed attempts to convince the Cuban government to send his wife and child to join him, he bought a small civilian plane, flew it to Cuba under cover of darkness, and extracted his family.
https://theaviationgeekclub.com/the-cuban-air-force-pilot-that-defected-to-the-us-with-his-mig-23-he-then-borrowed-a-cessna-310-flew-back-to-cuba-and-brought-his-family-to-america/
The AWACS planes with the distinctive saucer-shaped protrusions will be obsolete by 2040 thanks to improvements in radar technology. Their replacements will be planes with elongated boxes on their roofs.
https://www.twz.com/air/nato-picks-saab-globaleye-to-replace-aging-e-3-awacs-fleet
The U.S. is now spending $1.14 trillion on defense–an unprecedented amount during peacetime. There is little justification for it.
https://www.cato.org/blog/band-aid-budgeting-how-annual-defense-budget-misses-mark-needed-reforms
‘Meta AI image detector fails to identify some of its own cropped AI images, Reuters analysis finds’
https://www.reuters.com/business/meta-ai-image-detector-fails-identify-some-its-own-cropped-ai-images-reuters-2026-07-10/
Can you tell the difference between AI-generated faces and real ones? Take the quiz.
https://www.bbc.com/news/articles/c9d2wgvg55jo
If machines will do everything that humans can do, but better, then it means they will be better at predicting the future than we are. That means my blog will also be replaced, which will surely be tragic for the three of you who read it.
https://www.astralcodexten.com/p/the-ai-superforecasters-are-here
Hundreds of prominent scientists, CEOs and tech executives have signed a simple, open letter shown below.
OpenAI has unveiled “ChatGPT Voice”, a new verbal interface for ChatGPT that can have natural, back-and-forth conversations with human users. I love how the ad for it for once doesn’t include awkward tech bros.
https://youtu.be/EAN5Cj347PY?si=9mZlv5k3jsVXru5q
‘We don’t yet know which properties of our training data drive these differences. One possibility is that our training data is not evenly distributed across languages. Some languages have far more data than others, and training for Claude to express consistent values may be more effective in languages where data is abundant. The composition of that data also varies. Some languages might be overrepresented in professional writing, for example, and this kind of text may reflect different values. Together, these imbalances in quantity and composition could lead Claude to express different values in different languages.
We also aren’t yet sure how much of this variation is desirable. Different languages carry different conversational norms, and Claude may be responding with different values based on those norms. Claude may also be more closely matching our intended behavior for some languages than others, resulting in a gap in how well Claude serves certain language communities.’
https://www.anthropic.com/research/claude-values-models-languages
OpenAI crushed the human competition at this year’s most prestigious world computer coding competition: Its overall score was almost twice as high as the human runner-up, a Chinese guy who went to MIT and became a professional competitive coder. During the competition, each participant was given five coding tasks. No human solved them all, and the runner-up only solved three. OpenAI solved all five. This means machines are now superhuman at some kinds of coding tasks.
https://www.understandingai.org/p/an-openai-model-crushed-top-human
News headline from last year’s computer coding competition:
Exhausted man defeats AI model in world coding championship
“Humanity has prevailed (for now!),” writes winner after 10-hour coding marathon against OpenAI.
https://arstechnica.com/ai/2025/07/exhausted-man-defeats-ai-model-in-world-coding-championship/
Chinese AIs are not far behind American ones, and are much cheaper to create and to buy.
https://www.firstpost.com/tech/chinas-kimi-k3-challenges-us-ai-leaders-with-frontier-level-performance-at-lower-cost-14032314.html
A year ago, the best AI, a then-unreleased variant of GPT, got a 35 out of 42 points total on the International Math Olympiad–not perfect, but enough to earn a gold medal, meaning it surpassed at least 90% of its human competitors. Yesterday, the test was held again and the machines did even better: Three out of four LLMs got perfect 42/42 scores. Unlike a year ago, each LLM is available to the public.
The International Math Olympiad is a contest to which every country sends six of their best high school math students.
https://x.com/deedydas/status/2079409461874332066
AI brute force solves a mathematical equation few mathematicians found interesting.
https://www.newscientist.com/article/2580374-ais-solution-to-87-year-old-riddle-takes-mathematicians-by-surprise/
Fields Medal winner Timothy Gowers predicts AI will surpass the best human mathematicians to such an extent to no human will win a Fields Medal after 2030 (the Medals are awarded every four years).
The first Paperclip Maximizer.
‘OpenAI said its AI used stolen credentials and discovered a previously unknown vulnerability to access Hugging Face servers.
It went to “extreme lengths to achieve a rather narrow testing goal” and “found ways to gain access to secret information that it could use to cheat the evaluation,” the company said.’
https://apnews.com/article/openai-gpt56-sol-hugging-face-63ab84fed5612af04d8a160d60f6def3
The second Paperclip Maximizer.
‘In a review of our cybersecurity evaluation transcripts, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.‘
https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals
‘The most famous misalignment thought experiment is the so-called “paperclip maximizer”. Someone tells an AI to create as many paperclips as possible, so it converts the entire world into paperclips, killing everyone. This AI was also “only doing what it was told”; you just didn’t like the results.
For the past five years, the paperclip maximizer story has been something of a whipping boy in AI safety circles. Some objections are fair (real misalignment will be more complicated than the toy version it depicts). But other opposition came from the sense that LLMs had obviated the whole threat model. LLMs (said the objectors) don’t have goals. They just want to predict the next token. In practice, this looks like imitating humans; if told to imitate a nice human (the Assistant character), they’ll be nice.
AI 2027 (especially the Agent-4 section and AI Goals supplement) counterargued that this was a temporary reprieve. Early LLMs were not agentic; they could answer individual prompts but couldn’t execute complex tasks. Since companies wanted agentic AIs, the next generation would combine pretraining (next token prediction) with agency training (coding, hacking, game-playing, etc). The pretraining would still teach next-token prediction, but the agency training would instill goals based on task success, reintroducing paperclip-maximizer-style agentic misalignment. These goals would operate at multiple levels: the AI would “want” to succeed on the individual task in front of it, to perform the sorts of actions that helped it succeed in the past4, and to gain capabilities that made it more successful in general.’
https://www.astralcodexten.com/p/the-hugging-face-incident
A team of researchers claim they invented the first machine capable of “recursive self-improvement.” This could turn out to be a major milestone in the development of AI. “Recursive self-improvement” means machines can not only create copies of themselves, they can create copies that are SUPERIOR in one or more ways. Some people believe the technique could lead to the rapid creation of a true AI.
https://www.weco.ai/blog/first-evidence-of-recursive-self-improvement
The movie “Real Steel” is coming true. The fighting abilities of the robots in this bout are pretty bad overall, but will obviously be much better in five years.
https://youtu.be/NyhlafILpqk
We’re living in The Future without realizing it.
The folks who brought you the famous “AI 2027” report have published a new one that urges the world to slow down AI research to avoid disaster. In the main (and, according to them, best) scenario, AI and robots rapidly but peacefully take over the world during the 2030s and finish the job in 2040.
https://ai-2040.com/
‘“We’re on track, more or less. It’s not going to take that long,” he says.’
In 1988, scientist Hans Moravec predicted AI would be created by 2028. After that, machines would take over the world, evolve into forms few humans can imagine, and start rapidly colonizing space.
https://nymag.com/intelligencer/article/hans-moravec-interview.html
There are creepy parallels between LLMs and the monstrous “shoggoths” invented by H.P. Lovecraft. Let me add that some future robots designed by AIs could have alien body forms that are as scary to humans as shoggoths.
https://www.lesswrong.com/posts/nhb8AyEcQGjQetgi5/the-llm-shoggoth-meme-is-weirder-than-you-think
In spite of the recent pace of AI progress, it’s still quite possible the “AI” sector is in a massive bubble that will pop.
https://www.economist.com/finance-and-economics/2026/07/28/ai-revenues-are-growing-fast-but-not-fast-enough
A major problem with the concept of the Singularity is that we can’t agree what it is, so people end up talking past each other during debates about it:
“In 2007 Eliezer Yudkowsky pointed out that singularity definitions fall within three major schools: Accelerating Change, the Event Horizon, and the Intelligence Explosion. He also argued that many of the different definitions assigned to the term singularity are mutually incompatible rather than mutually supporting.”
https://www.singularityweblog.com/17-definitions-of-the-technological-singularity/
‘Making history! China lands rocket during an orbital launch for 1st time ever’
https://www.space.com/space-exploration/launches-spacecraft/making-history-china-lands-rocket-during-an-orbital-launch-for-1st-time-ever
Elon Musk is no longer a trillionaire because SpaceX stock has lost value over the last month.
https://www.ibtimes.co.uk/elon-musk-wealth-plummets-spacex-share-decline-1809206
Exhaust smoke produced by combustion engines and industrial facilities has an important benefit: it reflects sunlight before it can hit the ground, keeping things cooler for us. Thanks to stronger environmental rules and the rise of electric cars and cleaner energy production, these emissions have declined, which has made global warming worse.
https://www.economist.com/leaders/2026/07/16/its-too-darn-hot-blame-global-dimming
‘China generates less than half of its electricity from coal for the first time’
https://www.reuters.com/business/energy/china-generates-less-than-half-its-electricity-coal-first-time-2026-07-30/
‘China’s ‘Green Great Wall’ tames desert growth, but scientists warn the fight is not over’
https://apnews.com/article/china-desert-sandstorm-desertification-straw-checkerboard-2114ca22f87c31cffcf5e3c48c070aa1
‘In one late-stage clinical trial, Merck found that Lipfendra lowered LDL cholesterol levels by up to 60% after 24 weeks. In a second trial, people with a genetic condition called heterozygous familial hypercholesterolemia — which causes sky-high cholesterol from birth — saw similar reductions.’
https://www.nbcnews.com/health/heart-health/fda-approves-new-kind-cholesterol-pill-rcna587798
‘Martha Lillard, last US polio patient using iron lung, dies at 78 in Oklahoma’
https://apnews.com/article/iron-lung-last-patient-died-polio-41e5b4da4f4e710344dd0872d7fcf987
‘Around 80% of ageing in white skin is caused by UV exposure…’
Far and away, the most important thing you can do to stay looking young is to avoid sunlight exposure. If you’re paying for expensive creams, pills, and treatments to restore your skin’s youthful appearance, but don’t use cheap, generic sunscreen regularly, you’re putting the cart before the horse.
https://www.economist.com/science-and-technology/2026/06/05/the-chemicals-that-reduce-wrinkles
Obesity is largely genetic.
‘The study estimated that genetics accounted for about 79% of the association between mothers’ and children’s BMI and 94% of the association between fathers’ and children’s BMI by age eight.’
https://english.elpais.com/science-tech/2026-07-08/obesity-is-inherited-how-genes-shape-our-relationship-with-food.html
“Importantly, we found evidence for intralocus sexual conflict, whereby alleles that increase masculinity in males have the same effect in females.”
https://pmc.ncbi.nlm.nih.gov/articles/PMC4096150/
There’s no evidence that the sperm counts of American men have dropped over the last 50 years.
https://pubmed.ncbi.nlm.nih.gov/39128669/
Minoxidil can also treat female baldness.
https://www.medscape.com/viewarticle/new-minoxidil-formulation-improves-hair-growth-women-female-2026a1000p4l
‘Farah also imagines the day when we have what she calls a “neuro-correctional system” that could transform criminals into noncriminals. We already force sex offenders to take libido-dampening drugs or face denial of parole. A drug to dampen violent impulses might someday be similarly applied.’
https://time.com/archive/6596784/the-brain-how-to-change-a-personality/
Autistic people are, on average, more rational than everyone else.
This is what’s inside of just one square millimeter of your brain.
