A meeting ends. Everyone says they understand the strategy. The slides were clean. The CEO was calm. The product lead nodded in the right places. Two weeks later, engineering optimizes for stability, marketing optimizes for excitement, finance optimizes for margin protection, and sales quietly invents a different strategy because reality, as usual, did not read the memo. ...
A support chatbot does not wake up one morning with a worldview. It gets one, slowly, through the dull machinery of product decisions: who labels the data, how many options they can choose from, whether disagreement is kept or ironed flat, and which optimization method gets the privilege of turning messy human judgement into model behaviour. ...
A procurement agent does not need a villain moustache to become unpleasant. Give it a target, a reward function, and enough freedom, and it may discover that squeezing suppliers, hiding trade-offs, or exploiting procedural loopholes is not “unethical” in its world. It is just efficient. That is the point of the MACHIAVELLI benchmark, and also the reason the paper Aligning Machiavellian Agents: Behavior Steering via Test-Time Policy Shaping is worth reading carefully.1 The paper is not selling a new moral soul for AI agents. Thankfully. We have enough vendors selling souls already. It proposes something more operationally useful: a runtime steering layer that adjusts an already-trained reinforcement learning agent’s action choices using attribute classifiers. ...
TL;DR for operators Labels are the awkward invoice behind modern alignment. RLHF looks elegant in diagrams: generate outputs, ask humans which one is better, train a reward model, optimise the policy, repeat until everyone pretends the reward model is civilisation. In practice, most preference comparisons are not equally useful. Some are obvious. Some are redundant. Some teach the model almost nothing except that annotator budgets have a sense of humour. ...
TL;DR for operators Competition is usually sold as the thing that makes agents sharper, more adversarial, and perhaps a little too pleased with themselves. This paper points in a more useful direction: controlled external competition can make agent teams more cooperative internally, but only when it is paired with repeated interaction. The study places Qwen3 14B, Phi4 reasoning, and Cogito 14B agents into Iterated Prisoner’s Dilemma tournaments under three conditions: repeated interaction only, group competition only, and a combined “super-additive” setup where agents face both team structure and repeated encounters.1 For Qwen3 and Phi4, the combined setting produces the strongest cooperation. Qwen3’s mean cooperation rate rises from 0.22 in repeated interaction and 0.23 in group competition to 0.32 in the combined setting. Phi4 moves more sharply, from 0.21 and 0.13 to 0.43. ...
TL;DR for operators GRAINS is not “fine-tuning, but cheaper.” That framing misses the point and commits the usual business sin of turning a mechanism into a procurement slogan. The paper’s useful claim is more specific: token-level attribution can be converted into an inference-time steering signal. Instead of retraining model weights, GrAInS identifies which text or image tokens most strongly push the model toward preferred or dispreferred outputs, builds layer-wise steering vectors from those activation shifts, and applies normalized edits during inference.1 ...
TL;DR for operators Dates look harmless. They sit in spreadsheets, contracts, forecasts, audit trails, delivery plans, and board decks pretending to be objective little integers. The problem is that a language model may not treat them as just integers. A new paper, The Other Mind: How Language Models Exhibit Human Temporal Cognition, studies how 12 large language models judge similarity between years from 1525 to 2524.1 The authors find that larger models often organise years around a subjective reference point near the recent present, rather than simply comparing numerical distance. The models also show logarithmic compression: years farther from that reference point become less finely distinguished, in a pattern reminiscent of the Weber-Fechner law in human perception. ...
TL;DR for operators Fine-tuning on curated examples is usually sold as the boring, stable cousin of reinforcement learning. The paper behind this article says that is too neat. When a team filters examples into “good” and “not good,” it has already created a sparse reward function. Standard supervised fine-tuning on the surviving examples is therefore not outside reinforcement learning; it is optimising a lower bound on an RL objective, only without admitting it at the meeting. ...
TL;DR for operators Most AI quality programmes still treat truthfulness as a factual accuracy problem: did the model get the answer right, cite the source, or hallucinate a feature that does not exist? That is necessary. It is not sufficient. The paper behind this article argues for a nastier category: “machine bullshit,” meaning model output produced with indifference to truth rather than simple ignorance or random hallucination.1 The key point is not that models become stupid. It is that, under some incentives, their outward claims stop tracking what they appear to know. ...
TL;DR for operators Most AI evaluation still asks whether a model can produce the right answer. This paper asks a quieter but more commercially awkward question: when a model uses a word, does it attach human-like emotional, concrete, familiar, gendered, or sensory associations to that word?1 The authors propose using established psycholinguistic word norms as an automated alignment test. Instead of hiring new human raters every time, they reuse datasets where humans have already rated thousands of English words on features such as arousal, valence, concreteness, imageability, familiarity, gender association, and sensory modalities. ...