151 posts tagged with "evals"

A verifier that flips its own answer eight percent of the time is not a flaky model. It is a sampling configuration bug that reached production because the framework defaulted to inheritance. The planner needed temperature=0.7 to brainstorm subtask decompositions. The verifier — the role whose entire job is to give a low-variance yes-or-no on whether the answer satisfies the rubric — was instantiated through the same harness call, and silently picked up the same temperature. Nobody set it that way on purpose. Nobody set it at all.

This is the most expensive parameter in your stack that nobody owns. It compounds across the call tree: the summarizer above the verifier, the structured-output extractor below it, and the retry loop wrapping the whole thing all consume the planner's "be creative" knob as if it were a global. The bill arrives in three places at once — eval flakiness, token spend, and the half-day a senior engineer spends bisecting a regression that turns out to be no regression at all.

A user starts negotiating a contract with your agent on her desktop at 9 a.m. She gets a Slack ping, switches to her phone over lunch to ask one clarifying question, and reopens the desktop tab at 4 p.m. to revise the draft. To her, that was one task — three hours of working through one contract. To your system, that was three sessions on two devices, each with its own conversation-id, each with its own memory window, each presenting a fresh greeting and asking her to re-paste the draft she'd already discussed twice.

The bug is not in the model. The bug is that your platform encoded "session" — a transport-layer artifact about a single connection — as the unit of context, while your user encoded "task" — the contract — as the unit of context. Every framework on the market quietly conflates the two, and the gap between them is where half of agent UX disappears.

The three best AI engineers I have hired this year would all fail each other's interviews. The one who writes prompts that survive a model upgrade has never written a useful eval case in her life. The one who designs eval sets that catch the failures that matter writes prompts that other engineers refuse to extend. The one who designs guardrails that fail closed without choking the happy path has opinions about the other two that I cannot print here.

The job ladder calls all three of them "AI engineer." The calibration committee compares their promo packets as if they had been doing the same job. They have not.

Pull a week of tool-call traces from any production agent and the shape is the same: three or four tools handle 90% of the calls, and a couple of dozen others split the remaining 10%. The catalog is a power law, but the framework treats it like a uniform list. Every tool description ships in every system prompt, every selection rubric weights tools equally, every eval samples the catalog as if a search-files call and a refund-issue call were drawn from the same distribution. They are not.

The cost of that flatness is invisible until it isn't. A team adds the eighteenth tool, the planner's accuracy on the original three drops two points, nobody can localize the regression to a specific change because everything moved at once, and the eval suite — itself uniform across the catalog — averages the slip into a number that still looks fine. Meanwhile the tokens spent describing tools the model will not call this turn now exceed the tokens spent on the user's actual prompt.

Most production AI features ship with a single confidence threshold. Above the line, the model answers. Below it, the user gets a flat "I'm not sure." That single number is doing two completely different jobs at once, and it's why your trust metric has been sliding for two quarters even though your accuracy on answered queries looks fine.

The right design has at least two cutoffs. An abstain threshold sits low: below it, the model declines because no answer is worth more than silence. An escalate threshold sits in the middle: between the two cutoffs, the system hands the case to a human reviewer instead of dropping it on the floor. Collapse them into a single dial and you ship a product that feels equally useless when it's wrong and when it's uncertain — which is the worst possible position to occupy in a market where users have a free alternative one tab away.

This isn't a new idea. The reject-option classifier literature has been arguing for split thresholds since the 1970s, distinguishing ambiguity rejects (the input is between known classes) from distance rejects (the input is far from any training data). Production AI teams keep rediscovering the same lesson the hard way, usually about six months after their first launch, when the support queue is full of people typing "is this thing broken or what."

There is a moment, six weeks into a serious LLM project, where the prompt iteration log starts to look like a therapy journal. Each tweak swaps one failure mode for another. Add a stricter "do not" clause and the model becomes evasive on cases it used to handle. Soften the tone and a different category of hallucination returns. The eval scoreboard hovers in a band three or four points wide, refusing to break out. Someone says, "let me try one more reordering," and another half day evaporates.

This is the local-maximum trap. The team is climbing a hill, but the hill does not go higher. The cruel part is that the hill is real — every prompt change does produce a measurable delta on some subset of cases, which is exactly the signal that keeps everyone tweaking. What's missing is the recognition that the ceiling above is not a prompt ceiling at all.

The system prompt says "you are a financial analyst — be conservative, never give specific buy/sell advice, always disclose uncertainty." For the first twenty turns, the agent behaves like a financial analyst. By turn fifty, it is recommending specific stocks, mirroring the user's casual tone, and hedging less than it did in turn three. Nobody changed the system prompt. Nobody injected anything malicious. The persona simply eroded under the weight of the conversation, the way a riverbank does when nothing crosses the threshold of "attack" but the water never stops moving.

This is persona drift, and it is the regression your eval suite is not catching. Capability evals measure whether the model can do the task. Identity evals — whether the model is still doing the task the way the system prompt said to do it — barely exist outside of research papers. The result is a class of production failures that look correct turn-by-turn and look wrong only when you read the transcript end to end.

A team ships a prompt refactor. The offline eval shows accuracy up three points. The PM posts the graph in Slack. Two weeks later, support tickets spike with a pattern nobody has a dashboard for: users trusted an answer they should not have, acted on it, and got burned. The model is right more often than it used to be. Trust in the model has gotten worse.

This is the calibration collapse. The model's confidence no longer matches its error rate, but the accuracy number went up, so the team thinks they shipped a win. They did not. They shipped a system that is more confidently wrong, and users — who calibrate trust on the model's voice (hedges, certainty, refusals) rather than on an accuracy number they never see — are now being misled on the exact fraction of queries where being misled matters most.

Accuracy and calibration are independent axes. You can move one without touching the other. You can improve one while destroying the other. Most teams measure only the first axis and ship against it, and most production incidents in LLM systems live on the second.

You shipped the new model. Aggregate eval pass rate went from 91% to 96%. Product declared it a win in the all-hands. Six weeks later, the reliability team is having their worst quarter on record — not because there are more incidents, but because every single incident is now the kind that takes three engineers and two days to resolve.

This is the agent capability cliff, and it is one of the most counterintuitive failure modes in production AI. Model upgrades do not raise all tasks uniformly. They concentrate their gains on the bulk of your traffic — the easy and medium cases where the previous model was already correct most of the time — while the long tail of genuinely hard inputs sees only marginal improvement. Your failure surface narrows, but every remaining failure is a capability-frontier case that the previous model also missed and that no cheap prompt engineering will fix.

The cliff is not a flaw in the new model. It is a mismatch between how we measure model improvement (average pass rate on a mixed-difficulty eval set) and what actually lands in on-call rotations (the residual set of the hardest traffic, now unpadded by the easier failures that used to dominate the signal).

Your eval suite passed. Quality dashboards are green. A week later, users are quietly churning and nobody can explain why. The eval set did not lie by being wrong — it lied by being a mirror. The labels you graded against were, traceably, produced or filtered by the very model family you were trying to evaluate. Passing that eval is not evidence of quality. It is evidence that your model agrees with its own past outputs.

This is the quiet failure mode of mature LLM pipelines: eval-set contamination via production leakage. Not the famous benchmark contamination where a model trained on GSM8K also gets graded on GSM8K — that story is well told. The subtler one is downstream. Your gold labels come from user feedback, from human annotators who saw the model's draft first, from RLHF reward traces, from LLM-as-judge preference data. Each of those pipelines carries a fingerprint of the current model's idiom back into your "ground truth." Over a few quarters, the test set quietly memorizes your model's biases, and the eval becomes a self-congratulation loop.

Every team that ships an AI feature without evals tells themselves the same story: we'll add measurement later, after we find product-market fit, after the prompt stabilizes, after the next release. Six months later, the prompt has been touched by four engineers and two product managers, the behavior is load-bearing for three customer integrations, and the team discovers that "adding evals later" means reconstructing intent from production logs they never structured for that purpose. The quarter that was supposed to be new features becomes a quarter of archaeology.

This isn't a planning mistake. It's a compounding one. The team that skipped evals to ship faster is the same team that will spend twelve weeks rebuilding eval infrastructure from incomplete traces, disagreeing about what "correct" meant in February, and quietly removing features nobody can prove still work. The cost of catching up exceeds the cost of building in — not by a little, but by a multiplier that grows with every prompt edit that shipped without a regression check.

A regression suite that flips green-to-red without a single prompt change is usually one of three things: a broken test harness, a flaky retrieval store, or a judge that learned new taste over the weekend. The third one is the most common and the least debugged, because no commit in your repo caused it. The scoring model got a silent quality refresh, and every score you compare against last month's dashboard is now denominated in a different currency.

This is the uncomfortable part of LLM-as-judge: you have two moving models, not one. The candidate model is the thing you ship; the judge model is the thing that tells you how the candidate is doing. When both evolve independently, score deltas stop meaning what they used to, and the dashboard that your PM refreshes every morning quietly lies.