Ballmaid

[Ballmaid @Ballmaid_]

@scaling01 Holy shit this is actually fire

[Ballmaid @Ballmaid_]

@BrenBuilds @StoicYield This is such a large misconception.
5.5 reasons way less (about 2x less on most tasks on xhigh), which makes it equivalent cost-wise to 5.4.
With 5.5 being significantly more intelligent.
There is not a single reason to still use 5.4 over 5.5

[Ballmaid @Ballmaid_]

@MittaiApp @possiblyazure Zed mentioned 🔥🔥

[Ballmaid @Ballmaid_]

@kimmonismus Yes Herman miller is great.
To give you the tldr:
- HM Aeron
- HM Embody (I have this one, fixed all of my back/long hour comfortability problems)
- Haworth fern
- steelcase leap v2
- steelcase gesture

Best YT channel for this: Ahnesty
Generally I would recommend refurbished

[Ballmaid @Ballmaid_]

@ob35cne @Lexx79011 @Scivf4 All of machine learning is about finding a good enough abstraction for a problem to make it learnable for a neural network. We always work with assumptions to simplify reality for the task we want a model to achieve

[Ballmaid @Ballmaid_]

@ob35cne @Lexx79011 @Scivf4 Im not knowledgeable enough to make a good judgement here, so don’t take my answer as facts. Though how I see it, is that current DL models are already an abstraction of reality (and are required to be so). Our brains inherently learn inherently differently (neuroplasticity)

[Ballmaid @Ballmaid_]

@Lexx79011 @Scivf4 Because it’s literally just a mathematical function with a bunch of parameters.
It’s insane how far we can go with this though, but this has nothing to with consciousness.
The visualization is cool, but a bit misleading in relation to how simple a neural network is in principle.

[Ballmaid @Ballmaid_]

@0xnglif @sudoingX Oh yes I have seen that too, so yes you are absolutely right here. Though it‘s arguably still a few wasted tokens each time, which for each single new prompt (or back and forth/tool call) will add up over time of course

[Ballmaid @Ballmaid_]

@Kappaemme1926 @spacejaz8 @thsottiaux Claude limits are significantly worse, largely to be explained by the fact, that their models are not nearly as token efficient as GPT5.5

[Ballmaid @Ballmaid_]

@inqodo @llmdevguy Are you sure we are talking about the same GPT5.5?

[Ballmaid @Ballmaid_]

@R2Cdev_ It’s completely useless, like it’s not even possible to do such an realistic estimation since there are so many factors. Like number of active params, general architectural differences (this already makes comparison nearly impossible), different training data or quantization

[Ballmaid @Ballmaid_]

@amanhostednft Codex, more generous limits, GPT5.5 doesn’t reason unnecessarily, is more intelligent and code is great (also had Claude sub beforehand, in comparison imo GPT5.5 is better for code), access to GPT5.5-pro (that model is insane, I use it all the time for very difficult tasks/ideas)

[Ballmaid @Ballmaid_]

@John__Micheal @diegocabezas01 This idea of parameter estimation is anyways pretty useless, since what actually matters is the number of active params, not the total

[Ballmaid @Ballmaid_]

@thehenryinsf @koltregaskes Gemini 3.5 Flash is generally even more expensive than GPT5.5 because it wastes so many reasoning tokens. Gemini 3.5 Flash cost-performance is ratio is horrendous

[Ballmaid @Ballmaid_]

@TheSpacerr @theo @GaryMarcus Indirect positive bias shift towards OpenAI, even he would not be required to report positively about them

[Ballmaid @Ballmaid_]

@DissentFu @ByMSA21 @EoghanH @fuckyouiquit That’s absolutely not what I am claiming.
Our human brain is way more advanced and capable than LLMs.
It is just a fact, that LLMs surpass average human in some fields (like coding for instance), making it insanely useful for such tasks

[Ballmaid @Ballmaid_]

@DissentFu @ByMSA21 @EoghanH @fuckyouiquit This is the case if you actually do full task offloading. I’m using it to speed up study, do fast tests of research ideas of mine etc.
I agree, probably a large portion of people fully relying on AI will see decline.
If you use it correctly, I would argue it’s the opposite effect

[Ballmaid @Ballmaid_]

@DissentFu @ByMSA21 @EoghanH @fuckyouiquit I think the problem here might be that you did not use the newest models much yet. I would bet your perception would change if you would start doing so. I’m a student + researcher (at an Uni). LLMs have benefited me massively, I’m fascinated everyday what they are capable of