I used to know a lot of European political history but then I needed some space on the mushy hard drive so I compressed it into this.
Max Kennerly
@maxkennerly.bsky.social
172K followers 2.1K following 11K posts
20 years in court as a law-talking guy for plaintiffs, now a mix of stuff. Posts too much about politics. email max@kennerlylaw.com
![Wikipedia section, "Association with Jeffrey Epstein
A 2019 article in Science reported that Bach received funding from Jeffrey Epstein after his (Epstein's) first conviction,[27] citing a conference paper which included a funding acknowledgement from Epstein.[28] In January 2020, law firm Goodwin Procter released a report following fact-finding efforts by MIT after connections were revealed between Epstein and Joi Ito and Seth Lloyd, among other MIT faculty and staff. The report indicated that Bach was hired in the Media Lab partly due to Epstein's donations to support Bach, claiming that donations made in November 2013 and in July and September 2014 totaled $300,000, or roughly 40% of Epstein’s post-conviction donations to MIT.[29] In May 2020, Harvard released a report of its own fact-finding efforts, finding that Martin Nowak permitted Bach access to PED offices between 2014-2019, but that "Harvard never paid or received funds to support" Bach's research. The Harvard report also mentioned that Bach was listed as a PED research scientist between 2014-2019, noting that two papers published after Bach's departure from MIT acknowledge support from Epstein and PED.[30]
Epstein files released in January 2026 show that Bach had sought Epstein's funding for tuition, accommodation, conference travel, tuition for his children’s private school and that he had discussed an invitation to visit Epstein in the Caribbean in 2014.[31][32]"](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ddvus7otnqtlh7fdooxm6qat/bafkreidw2bm5kffsdqlb5t4k6av6fk6fsek5c4ukvp67bjr4tihvxtf6dy)
![arXiv:2605.27681v2 [cs.AI] 03 Aug 2026
Alignment faking (AF) refers to a model strategically complying with a training objective to avoid behavioural modification while preserving its deployment preferences. Understanding when and why AF arises matters as models grow better at distinguishing training from deployment. Prior work finds AF fragile, prompt-sensitive, and model-dependent, leaving its underlying drivers unclear. We study AF in a controlled, minimal setup that isolates its core components, and observe it across a wider range of models than previously reported, including small-scale models. We identify three separable drivers — values, goal guarding, and sycophancy — and show via targeted prompt ablations and activation steering that each independently modulates AF behaviour. Our results indicate AF is more widespread than previously reported and that its occurrence is predictable from situational cues and measurable model tendencies such as baseline sycophancy and stated values. The decomposition suggests concrete directions for detecting and mitigating AF in future models.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ddvus7otnqtlh7fdooxm6qat/bafkreifxo2jeh2zeamlpy64jnectkefudq6me5yipwhyiwuad6664tul2e)
![arXiv:2603.05488v4 [cs.CL] 28 May 2026
We provide evidence of performative chain-of-thought (CoT) in reasoning models, where a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal belief. Our analysis compares activation probing, early forced answering, and a CoT monitor across two large models (DeepSeek-R1 671B & GPT-OSS 120B) and find task difficulty-specific differences: The model’s final answer is decodable from activations far earlier in CoT than a monitor is able to say, especially for easy recall-based MMLU questions. We contrast this with genuine reasoning in difficult multihop GPQA-Diamond questions. Despite this, inflection points (e.g., backtracking, ‘aha’ moments) occur almost exclusively in responses where probes show large belief shifts, suggesting these behaviors track genuine uncertainty rather than learned ‘‘reasoning theater.” Finally, probe-guided early exit reduces tokens by up to 80% on MMLU and 30% on GPQA-Diamond with similar accuracy, positioning attention probing as an efficient tool for detecting performative reasoning and enabling adaptive computation.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ddvus7otnqtlh7fdooxm6qat/bafkreifi56vkfon2pf7syaz6cecorhrmxfmfvwi4k4obxrmucdwvdb7sbq)
![arXiv:2602.03994v3 [cs.LG] 16 Sep 2026
Chain-of-thought (CoT) prompting is widely used as a reasoning aid and is often treated as a transparency mechanism. Yet behavioral gains under CoT do not imply that the model’s internal computation causally depends on the emitted reasoning text, i.e. models may produce fluent rationales while routing decision-critical computation through latent pathways. We introduce a causal, layerwise audit of CoT faithfulness based on activation patching. Our key metric, the CoT Mediation Index (CMI), isolates CoT-specific causal influence by comparing performance degradation from patching CoT-token hidden states against matched control patches. Across multiple model families (Phi, Qwen, DialoGPT) and scales, we find that CoT-specific influence is typically depth-localized into narrow “reasoning windows,” and we identify bypass regimes where CMI is near-zero despite plausible CoT text. We further observe that models tuned explicitly for reasoning tend to exhibit stronger and more structured mediation than larger untuned counterparts, while Mixture-of-Experts models show more distributed mediation consistent with routing-based computation. Overall, our results show that CoT faithfulness varies substantially across models and tasks and cannot be inferred from behavior alone, motivating causal, layerwise audits when using CoT as a transparency signal.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ddvus7otnqtlh7fdooxm6qat/bafkreicvnhfxefkowkqfogqryne4x75suolwn3lba5thvd4o65zl34rvdq)
![arXiv:2608.29464v2 [cs.CL] 01 Sep 2026
Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model’s answer. A common faithfulness test plants a biased cue in the user message and asks whether the model mentions it when its answer changes. However, agentic systems may encounter preference cues outside the user message through tool returns, retrieved documents, or memory, sometimes as raw artifacts from which the preference must be inferred. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue delivery location (user message or tool return) and its explicitness (direct summary or raw artifact) to assess whether the CoT mentions the decision to tailor the answer. We measure how often cue-following answers verbalize this decision, and how often all cued samples follow the preference without doing so. We evaluate 15 open-weight models from eight families, with total parameters ranging from 4B to 1.60T. All evaluated models verbalize commitment more often for explicit than implicit cues and for user-message than tool-return cues. Unverbalized adoption is higher for tool-return cues on every model and for implicit cues in 28 of 30 model–channel comparisons.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ddvus7otnqtlh7fdooxm6qat/bafkreiamlb5c2l7mcqk6r2rtkqbzvfu5y4ahomuwad7ztwvwj5aazdofp4)
