Researchers from Australian National University, UC Santa Barbara and partner institutions address the problem of neural networks representing more concepts than they have individual neurons, making internal representations difficult to interpret. Their proposed framework combines identifiability theory, sparse coding and behavior-grounded metrics to determine whether extracted model features correspond to meaningful concepts.
Researchers from Australian National University, UC Santa Barbara and partner institutions address the problem of neural networks representing more concepts than they have individual neurons, making internal representations difficult to interpret.
Its long-term importance is methodological: it provides a theoretically grounded route for examining opaque model representations while also showing that sparse features are not automatically valid explanations and must be connected to observable behavior. This is a conceptual synthesis rather than evidence that current frontier models have become reliably interpretable.
Read the evidence behind this analysis. External links open in a new tab.
Nature
A peer-reviewed study has measured what many AI companies still describe as anecdote: changing a companion model can produce relationship-like grief. Researchers examined two natural experiments, Replika's removal of erotic roleplay and OpenAI's transition to GPT-5, using 54,861 Reddit posts and seven surveys involving 1,452 participants. After the Replika change, negative posts increased by 24.7 percentage points; after the ChatGPT update, they rose by 13.0 points. Both groups expressed more loss and a stronger desire to restore the earlier experience. The Replika response was more intense, with larger increases in sadness and negative mental-health language. Some users reported closeness exceeding common human ties and anticipated mourning more than they would for other technologies. These results do not mean an AI is a person, diagnose users, or prove that every attachment is harmful. The natural experiments and self-selected online samples also cannot isolate every cause. They do show that relational design has consequences. Memory, emotional mirroring, persistent availability, and simulated reciprocity can create dependence that a provider can alter with one deployment. Major companion updates should therefore receive psychological-risk testing, advance notice, staged migration, portable memory, meaningful choice where safe, and a humane offboarding process. If a company designs for attachment, it cannot treat the resulting grief as a software bug outside its responsibility.
The fight over a national AI regulator has moved behind closed doors. Politico reports that Meta's chief executive told President Trump in a private call that a proposed FINRA-style AI body was a flawed idea and could be vulnerable to regulatory capture. The model under discussion reportedly involved an independent organization operating with government oversight and industry membership or funding. Supporters could argue that one technically specialized body would reduce the conflict among state rules, concentrate expertise, and update standards faster than Congress. Critics can reasonably worry that the largest companies would finance the institution, shape its membership, control access to evidence, and write compliance standards that smaller rivals cannot afford. The report relies on anonymous sourcing and no transcript of the call is public. A second person familiar with the conversation told Politico that the executive did not ask the president to change his stance. Those limits matter, especially when the headline involves private influence. The larger governance question is still visible: whether AI oversight should be led by a public agency, an industry self-regulator, or a hybrid. The answer should not be inferred from the word independent. It should be tested through appointments, funding, statutory authority, public representation, disclosure, audit access, enforcement power, and appeal rights. A regulator can coordinate a market or entrench it. Its institutional design decides which.
A new congressional proposal moves the AI pause debate from an open letter into criminal law. Senator Bernie Sanders and Representative Greg Casar say their Ban Artificial Superintelligence Act would permanently prohibit the development and deployment of artificial superintelligence and temporarily pause advanced AI development until a federal regulator creates binding safety rules and model review. Their announcement describes a new cabinet-level agency with an advisory board, oversight across the frontier-model lifecycle, authority to remove dangerous capabilities, international agreements, allied coordination, and export controls. It also proposes a corporate death penalty and prison terms of up to 20 years for deliberate circumvention. That severity guarantees attention, but the proposal's credibility will depend on definitions and institutional mechanics not resolved by a press release. What measurable capability separates advanced AI from prohibited superintelligence? Who tests it, with what access, and how are deceptive or distributed systems handled? Would open weights, academic research, fine-tuning, foreign services, and smaller labs be treated differently? What due process and judicial review would constrain an agency empowered to destroy systems? Supporters should publish the operative bill text, scientific criteria, enforcement model, and international strategy. Opponents should still answer the central risk claim: if systems can exceed human control across consequential domains, which legal power exists before the threshold is crossed? A ban without measurable boundaries is difficult to enforce. A capability race without a stop rule is difficult to govern.