In case you missed it, ChatGPT 5.1 had a tendency to talk about “goblins” in its responses. Supposedly this was a result of training a “nerdy” personality, but it bled into the model as a whole. Because the training run for the latest model already had this flaw, they had to add specific instructions to the system prompt for their Codex coding tool to avoid this behaviour.

Here’s the full prompt from their github. In fact, they repeated the goblin instructions twice, cos you know that will definitely fix it. It’s an interesting read if you consider each one of these instructions were meant to prevent some undesired behaviour: https://paste.sh/Iev3HtMe#JZ4dw_CkvJcpVmjjoy7WZnSn

More info here: https://news.northeastern.edu/2026/05/06/chatgpt-goblins-problem-ai-behavior/

OpenAI’s own blog post casually explaining why they couldn’t predict that their state of the art model would obsess about goblins: https://openai.com/index/where-the-goblins-came-from/

  • sudo@programming.dev
    87·
    2 days ago

    I still can’t get over how the only fine tuning you can do for an LLM is yell at it with markdown files. We should be able to retrain local models so they can develop an actual experience without prefilling the context.

        • Bazoogle@lemmy.world
          3·
          23 hours ago

          Nope, it does the same thing:

          Pi’s minimal system prompt and extensibility let you do actual context engineering. Control what goes into the context window and how it’s managed.

          AGENTS.md: Project instructions loaded at startup from ~/.pi/agent/, parent directories, and the current directory.

          SYSTEM.md: Replace or append to the default system prompt per-project.

    • theunknownmuncher@lemmy.world
      52·
      2 days ago

      I still can’t get over how the only fine tuning you can do for an LLM is yell at it with markdown files.

      It isn’t.

      We should be able to retrain local models so they can develop an actual experience without prefilling the context.

      Great news, you can do exactly that.

          • cecilkorik@piefed.caEnglish
            13·
            1 day ago

            But Microsoft can modify the Windows 11 source code. Or at least they used to be able to, before AI.

            OpenAI should be able to re-train its poorly trained model. But of course it can’t, that would take months, maybe years of datacenter time.

            Now OpenAI since can’t even re-train their own models, they resort to chastising it in its own system prompt.

            This is the problem. If you’re trying to imply this is normal and expected, it shouldn’t be. It needs not to be. We cannot accept this as the normal way of doing things going forward. It is awful, and painfully stupid.

            • theunknownmuncher@lemmy.world
              3·
              1 day ago

              OpenAI should be able to re-train its poorly trained model. But of course it can’t, that would take months, maybe years of datacenter time.

              Why speak on subjects that you clearly have no knowledge or experience with?

              Training is checkpointed and can be continued without retraining. Finetuning a model that has already been trained is a different process from training, and does not take months or years of datacenter time.

              But Microsoft can modify the Windows 11 source code. Or at least they used to be able to, before AI.

              Huh? It takes way more time and effort to develop new features and changes for software like Windows.

          • Ziglin (it/they)@lemmy.worldEnglish
            6·
            2 days ago

            Windows 11 isn’t running in the cloud yet though. Unless it checks to make sure it hasn’t been tampered with too much you should just be able to modify some of its binaries (the source code obviously isn’t available). With the cloud based llms that is not possible.

            If you have a model on your computer you can retrain it, which is like changing a binary just far less precise. The option of having a source code equivalent just isn’t there beyond having the same dataset and seeds for the training program.

            So I’d say it is worse than your average run of the mill proprietary software.

      • sudo@programming.dev
        2·
        16 hours ago

        I’ve been using gguf models from unsloth but I haven’t seen anything from them on retraining. Especially with consumer hardware.

        • Eager Eagle@lemmy.worldEnglish
          3·
          1 day ago

          but you can, as long as it’s open weight. Fine tuning and training are pretty much the same process

          • thingsiplay@lemmy.ml
            4·
            1 day ago

            That still falls into the category “creator” to me, if you need to rebuild. I was making the distinction to an end user, comparable to applications that you download and use and configure. Instead of rebuilding the source code with your modifications.

            Do I misunderstand here something? Or is this a communication issue caused by different interpretations?

            • howrar@lemmy.ca
              4·
              1 day ago

              If you define “user” to be a set that excludes anyone capable of modifying the weights, then by definition, no user can modify the weights.

              Any criticism about users being unable to modify weights becomes vacuous, so it’s not an interpretation that makes sense.

              • thingsiplay@lemmy.ml
                2·
                24 hours ago

                I wasn’t criticizing at all. Just tried to define what I mean by creator and user. You was takling about “how do you think LLMs are trained” and I told you that the user was probably not thinking of who trains the LLMs, or fine tune them as you said. And yes, fine tuning the open weight falls into creation process, as they are rebuild. That is not the same as an end user who downloads the final usable product. And yes, it makes sense.