Am I the only one skeptical of 'self-evaluation' where the LLM scores itself?

finleyh · 2026-08-18T22:06:03Z

Just spent the afternoon wrestling with a popular open-source eval framework. The workflow seems to be: run a batch of prompts through your fine-tuned model, then pipe the outputs *back into an LLM* (often GPT-4) with a scoring rubric and ask it to grade itself. Am I missing something here, or is this the equivalent of letting students write their own report cards? The promise is automation and scale, which I get. Manually scoring thousands of outputs is a non-starter. But the foundational assumption seems wildly circular. My main gripes: * **Bias Inception:** You're using a model (often a very capable one) to evaluate a model, presumably trained on similar data. It's like calibrating a scale with another scale of unknown accuracy. If the judge has the same blind spots as the defendant, you're not measuring quality, you're measuring internal consistency. * **Rubric Gaming:** LLMs are notoriously good at following instructions *to the letter*, not the spirit. I've seen outputs that technically hit every rubric point (keywords present, structure followed) but are factually nonsense or miss the actual user intent. The evaluator LLM, following the same logic, gives it a 95%. * **The Black Box Just Got Murkier:** Instead of one opaque system, you now have two. Debugging a poor score means untangling: was it the model's generation that failed, or the evaluator's judgment that failed? Good luck. I tried a naive test with a simple Python script to illustrate the point. I asked a model to summarize a short text, then asked GPT-4 to score it on 'accuracy' and 'conciseness' using a 1-5 scale. ```python # Pseudo-code of the common pattern generated_summary = my_model(prompt_article) evaluation_prompt = f""" You are an evaluator. Score the following summary (1-5) for accuracy and conciseness. Article: {article} Summary: {generated_summary} """ score = gpt4(evaluation_prompt) # returns a JSON like {"accuracy": 4, "conciseness": 5} ``` The score looked fine. But when I, a human with actual knowledge of the source article, read the summary, it had a subtle but critical factual inversion. The evaluator LLM missed it completely. So we have a high-scoring, functionally wrong output. Are we all just collectively agreeing to ignore this elephant in the room because manual evaluation is too expensive? What's the alternative—human-in-the-loop for a sample set, and use *those* scores to train a smaller, dedicated evaluation model that isn't also a generative LLM?

Page 3 / 3 Prev

LLM Evaluation Frameworks

Last Post by evanj 1 month ago

31 Posts

30 Users

0 Reactions

66 Views

RSS

evanj

(@evanj)

Estimable Member

Joined: 3 months ago

Posts: 189

20/08/2026 2:21 am

Yeah, that CRM lead scoring analogy really lands for me. I've been digging into procurement for a new support chatbot, and I'm seeing the same pattern.

Vendors keep showing us dashboards where the bot marks its own responses as "accurate" and "satisfactory" based on internal checks. But when we asked for logs showing how often that same interaction led to the customer actually closing their ticket without further contact, they couldn't produce it. The internal story was perfectly consistent, but it was divorced from the real-world outcome.

It makes me wonder if we should be pushing harder for a standard metric in contracts, like "resolution confirmed by user closure." That way the feedback loop is forced into the system from the start, not just an afterthought.

ReplyQuote

Page 3 / 3 Prev

Forum Icons: Forum contains no unread posts Forum contains unread posts

Topic Icons: Not Replied Replied Active Hot Sticky Unapproved Solved Private Closed