Online Now: The Umwelt Representation Hypothesis: rethinking Universality https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(26)00162-2?rss=yes&utm_source=dlvr.it&utm_medium=mastodon
It is good to acknowledge that we don't know and not form strong opinions or narratives.
This Fed chart is one of those.
Anything past ±1% annually seem very unlikely to me , but I don't know that, because we have not discovered the best uses for current LLM-era AI, and other completely different developments on their way in robotics and world models are likely to be more transformative than what we have today.
Advances in AI will boost productivity, living standards over time
https://www.dallasfed.org/research/economics/2025/0624
AI: Considerations for people who make decisions
https://berthub.eu/articles/posts/ai-for-decision-makers/
Because they must uphold the law of large numbers.
A brief history of Luddism
https://www.economist.com/finance-and-economics/2026/07/23/a-brief-history-of-luddism
"“Luddite” has become a slur for technophobic stick-in-the-muds, the whingers who stand in the way of progress and want the clock to be turned back to a better time. [...] Economists and historians are kinder to the Luddites. The weavers were selective in their smashing, breaking only those machines that made what they perceived to be lower-quality goods. They were not ignorant about the potential benefits of automation. Nor were they the broad-based movement of the destitute that later lefty historians claimed; they were a small bunch of relatively well-off workers, often keen to preserve the demand for skills gained through a long apprenticeship. And it is not clear whether they were right or wrong about the immediate effects of automation. "
Why was Iran Attacked infographic from https://www.slow-journalism.com/
"The most important single statistic about the drug industry is the clinical failure rate" (percent of drugs that fail human clinical trials): 91% over the past two decades.
About 77% of failures are attributed to biological factors like inadequate clinical safety and efficacy, which are parameters typically aquired in preclinical animal studies.
The suggested solution comes from New Approach Methodologies (NAMs), like:
- in vitro models (cell and tissue-based) including
2D cell cultures, 3D spheroids and organoids, organ-on-chip systems
- In silico models (computational toxicology)
- Omics: transcriptomics, proteomics, metabolomics
- In chemico methods: cell-free, chemical- or biochemistry-based laboratory assays
- High-Throughput Screening: automatic robotic laboratories.
- Framework Integration: integrating all above and using conceptual causal-chains from molecular integration to adverse effects.
Clinical Failure Rates Over the Decades: Yikes
https://www.science.org/content/blog-post/clinical-failure-rates-over-decades-yikes
Need for NAMs: A systematic evidence synthesis revealing over half a century of drug development failure
https://www.sciencedirect.com/science/article/pii/S3050620426000448
I've found that often modern AI systems are better at longstanding mathematical conjectures than at newly posed mathematical questions. This makes me wonder if the system is doing the (awesome) work of bringing together the last many decades of human insights on open questions.
If so, then I do very much hope we can get these systems to properly attribute according to mathematical norms at some point.
It is counterintuitive, also, because humans are the opposite way. But that is good. That again shows how nicely humans and AI systems can complement each other if we build with that in mind.
Rigid rules can protect cooperation when motives are opaque.
Why human societies adopt rigid moral rules: The efficiency–robustness trade-off
https://www.pnas.org/doi/10.1073/pnas.2535467123
Rigid norms function as social technologies that trade efficiency for robustness against strategic defection, extending the domain of cooperation by eliminating exploitable ambiguity.
Flexibility maximizes welfare when there is trust, rigidity preserves cooperation when trust is fragile. This explains why rigid rules are common when others are perceived as unreliable.
European Password Manager Shares Origins and Updates with State-Certified Russian Firm https://www.occrp.org/en/investigation/european-password-manager-shares-origins-and-updates-with-state-certified-russian-firm
Passwork, a Spain-based password manager used by European government agencies and universities, shares technological ties with a Russian counterpart — an arrangement that experts say poses a state-level security risk.
Good News Everyone!
A large-scale randomized controlled trial in France shows huge improvement in teenagers baseline information inequality (45-70%), news consumption (+0.49 to 0.60 sd) and media literacy skills (+0.08 to 0.26 sd) with this simple trick: Give them free subscription to quality newspaper.
Making Teenagers Read Newspapers. A Nationwide Experiment in French High Schools
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7042478
Podcast and transcript describing the study: https://cepr.org/multimedia/making-teenagers-read-newspapers
#journalism #media #economics #socialScience #politicalEconomy #newsmedia
Making scientific knowledge free for all https://theconversation.com/making-scientific-knowledge-free-for-all-286857
The three main approaches are Green OA, Gold OA and Diamond OA:
- Green OA : publishing in a traditional journal and then depositing an accepted but unformatted version of the manuscript in an Open Access archive or an institutional repository
- Gold OA: final article immediately accessible to everyone, but involves publication fees amounting (i.e. Article Processing Charge, APC) to several thousand euros, paid by the authors or their institutions
- Diamond OA: allows authors to publish content for free and gives readers free access, thanks to journals that are backed by scientific communities, university libraries or non-profit organisations.
2/2:
There are functional and structural similarities between the J-space and theories of conscious access in humans, like the global workspace model.
Notable differences from human cognition: https://transformer-circuits.pub/2026/workspace/index.html#discuss-differences
Relationship to theories of consciousness:
https://transformer-circuits.pub/2026/workspace/index.html#discuss-conscious
1/2
A global emergent workspace in language models (J-space) https://www.anthropic.com/research/global-workspace
J-space wasn’t designed or programmed. It emerged on its own during training process. J-space has a number of unique properties:
- Claude can report on these representations. If you ask Claude what it's thinking about, it will tell you what’s in the J-space. Non-J-space representations are less reportable.
- It can also modulate them on request. If you ask Claude to think about something, or solve a problem silently in its head, it will light up the appropriate patterns in its J-space. By contrast, it has trouble modulating patterns not in the J-space.
- Claude uses its J-space for internal reasoning. If you ask Claude to solve a problem that requires multiple steps, the intermediate steps will light up in its J-space, even when it doesn’t say them out loud. These J-space patterns causally mediate its performance in such tasks, despite being smaller in magnitude than other representations.
- Representations in the J-space can be used flexibly for many tasks—for example, once “France” has lit up in Claude’s J-space, the model can recall its capital, or its national currency, or the continent it belongs to.
- However, despite its important role, the J-space is not involved in most of what a language model does—speaking fluently, recalling simple facts, using correct grammar, etc. In experiments where we prevented Claude from using its J-space, it still interacted normally, but lost its higher-order cognitive functions.
Full Paper:
https://transformer-circuits.pub/2026/workspace/index.html
External commentary https://www-cdn.anthropic.com/files/4zrzovbb/website/cc4be2488d65e54a6ed06492f8968398ddc18ebe.pdf
#consiousness #cogneuro #LLM #AI #cogneuro #cognitive #cognitiveNeuroscience
Problem: Self-improving agents are only as strong as the evaluator scoring them, they eventually stall.
Red Queen hypothesis in evolutionary biology: species must constantly adapt, evolve against similarly ever-evolving opposing species (co-evolution).
The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators https://arxiv.org/abs/2606.26294
In Red Queen Gödel Machine (RQGM) the evaluator is part of the search, letting agents and the criteria that judge them co-evolve.
This is very preliminary work, but the idea seems interesting.
https://www.climate.us/ formerly Climate.gov, is a effectively a copy of climate.gov.
A group of former NOAA employers who had been fired by Elon Musk's DOGE started a new climate science website.
A Philosophical Look at System Dynamics
https://www.youtube.com/watch?v=XL_lOoomRTA
Dartmouth College, Hanover, New Hampshire, Spring of 1977. In this lecture, Donella Meadows takes on a more philosophical concept. How can we bring ourselves to be aware of the assumptions we make as systems thinkers? She asserts that models are a set of assumptions. Donella Meadows defines some of these system dynamics assumptions (such as causal relationships and feedback loops) in this video.
(causal linkages, feedback loops, rates and levels, and structural
behavioral relationships)
https://en.wikipedia.org/wiki/Donella_Meadows
Sara Wickström awarded the Körber Prize 2026
https://koerber-stiftung.de/en/projects/koerber-european-science-prize/all-prizewinners/2026-sara-wickstroem/
Finnish physician and cell biologist Sara Wickström has been awarded the €1 million Körber European Science Prize for discovering how cells sense the physical world around them. Her research showed that cells can “feel” forces such as pressure and stretching and relay these signals all the way to their DNA. In doing so, physical forces can switch genes on or off, helping determine how cells develop and how tissues heal after injury. The discovery opened a new field of research and could eventually lead to new ways of treating cancer, organ fibrosis and other age-related diseases.
The article explaining it "The hidden senses of cells"
https://koerber-stiftung.de/en/projects/koerber-european-science-prize/all-prizewinners/2026-sara-wickstroem/#s54793
Writing code versus shipping code: Productivity effects across generations of AI coding tools
https://cepr.org/voxeu/columns/writing-code-versus-shipping-code-productivity-effects-across-generations-ai-coding
"In software, AI's most advanced application domain, a 180% gain in coding activity becomes a 30% gain in shipped releases and, so far, no detectable gain in usage. Growth projections that apply task-level estimates without adjusting for downstream bottlenecks will therefore overstate AI's near-term impact."
... "Human bottlenecks in reviewing, integrating, testing, and releasing code, together with scarce consumer attention, help explain why large task-level gains have not yet produced a software output boom. "
Humans are the only animals with chin. The chin isn’t just the lower part of your face: It’s a specific term for that little piece of bone extending from the jaw. https://www.smithsonianmag.com/smart-news/chin-stroking-mystery-why-are-humans-ony-animals-with-chins-180957997/
The most likely explanation is that chin is a a spandrel, a phenotypic trait that is a byproduct of the evolution of some other characteristic, rather than a direct product of adaptive selection, https://en.wikipedia.org/wiki/Spandrel_(biology)
Chin spandrel is probably related to patterns of facial reduction in humans during the Pleistocene,
Alternative adaptive hypotheses relate chins to speech, mastication, and sexual selection.
🧠 What can a single neuron compute?
A new #bioRxiv #preprint introduces #TwinProp, using a differentiable digital twin to optimize synaptic strengths and locations in a detailed layer 5 #pyramidal #CellModel. One #neuron classified images and speech and even solved 10-bit parity with 91% accuracy. The key: distributed #dendritic nonlinearities, especially #NMDA-dependent integration:
📄 https://doi.org/10.64898/2026.06.08.730984
#CompNeuro #Neuroscience #Dendrites #NeuroAI @albertcardona https://mathstodon.xyz/@albertcardona/116764629463190653