Shadow C | 🍔Arbiter🍕 🌭of🥪 🌮Sandwiches🌯

@michael_waples @0xdippo @thsottiaux There is a "waste" of tokens in the previous cycle, yes. But you also get an additional fresh cycle. So it's more like -80 and +100. Notice, this is why, when total days are preserved, the one with the extra reset - logically - ends with a higher cycle number than without.

@ShadowC5 @0xdippo @thsottiaux That comparison still does not fix the issue, because the loss happens at the transition. In no situation can you get them back. If they change the reset day you lose if you were going to use the tokens before the reset.

Hi. Over the last 24 hours we had three separate small incidents that affected Codex reliability. Those are three too many and we are taking active steps for them to not reproduce. I have reset usage limits for Codex across all paid plans. May the tokens flow again.

@michael_waples @0xdippo @thsottiaux No reset case Total experiential tokens = 80 (thursday, cycle 1) + 20 (friday, cycle 2) + 80 (next thursday, cycle 2) = 180 Reset on thursday Total experiential tokens = 80 (thursday, cycle 2) + 20 (friday, cycle 2) + 80 (thursday, cycle 3) = 180

@michael_waples @0xdippo @thsottiaux The above is about "accessible" token, as in what you can maximally use given the same timeline. But how about "experientially"? Say we have a pattern of using 80 tokens on thursday and 20 on friday post-reset.

@athiestboi It's completely fair to ask others to make their case if they so choose. Atheism being the default position, is more about one's own internal evaluation; and is not applicable in situations where ones tries to convince another of a position.

At this point Christians are just retard maxing.

@ivanburazin 400s under 30 here. Though my albumin and shbg were so high that effective free T was as good as being just in the 100s. Wild stuff because I only checked when gyms were all closed during covid lockdown, and I "had nothing better to do".

I've been on testosterone replacement for two years now. Once I figured out what was going on and got on a low dose, everything improved. My levels were at 301 when I found out. The healthy range for a man under 40 is anywhere between 500 and 1000. 301 is dangerously low. Everything is harder when your levels are that low. Sleep, focus, energy, building muscle, keeping fat off, all of it. Whenever I tell this to guys, the response I get always boils down to one of two things. Either "I've never checked my levels in my life..." or "I'm a man, I don't have that problem!" - which I always find hilarious. Women periodically visit doctors and get their hormones checked because they have structural reasons to. Men don't, so they never find out.

@TakeThatNurses @LinchZhang You actually can't tell if he used diameter or radius here, because radius is half diameter, so 1.5 is the remaining constant after eliminating both radius/diameter and/or the conversion factor of 0.5/1.

@LinchZhang Why are you using diameter instead of radius? I get that the answer comes out the same in terms of which is more pizza, but I’m just wondering why you made that choice for the area of a circle

1.5^2 = 2.25 2.25 > 2 QED All the constants are unnecessary here

@spirodonfl @cmuratori The very least, I would expect college grads to be well aware on some level of the social harms caused by social media, increasing polarisation due to media bubble built by recommendation engines, etc.

@spirodonfl @cmuratori I agree that AI is the clearest mirror right now, but I wouldn't go so far as to say that the harms of Big Tech is something that the audience (i.e. college grads, not random Gen Alphas) are that oblivious to either. Not necessarily Meta buying VPN companies, but at...

An article on @cmuratori 's latest video if anybody cares. Maybe I'll do a video on it too, I dunno. Title: A different perspective of Casey Muratori's 'But it happened' video spirofloropoulos.com/techbytes/issu…

@thsottiaux Also, it is much easier to reach over the desk to ask the friend, how he got the model or harness to work in the way he likes it best. Benchmarks don't give that. They just give general scores for untuned models. There is no next steps or what to try next.

@thsottiaux Friends are an underrated but big factor. It is inevitable that your closest connections are so, because you have something in common - what you work on, how you work, think, or communicate, etc. Friend who has success often is a better predictor of own success than benches.

Do you still trust benchmarks or do you just listen to your friends? What makes you try a new model?

@Sabre9186 @cmuratori I think the problem with "make use of this hindsight" is highlighted by another part of the video - "how? what? why?" There was no post mortem about why the bad thing happened. No advice on what to take special note of. Not even a brief hint of decision pitfalls.

@cmuratori Schmidt's using the phrasing to reflect *intention*, not culpability. He didn't target these effects, but they were real externalities of his choices. He's telling the next generation that he was naive, and to make use of this hindsight when they have similar responsibilities.

[1/2] I don't normally make commentary videos, but after seeing the entirety of Eric Schmidt's University of Arizona commencement speech, I felt like there was a lot more going on than just "CEO mentions AI, gets booed". So I made a video to explain what upset me about it.

@tomieinlove @kaywisehero @Perrid13 You can have 2-sided polygon, and there exists 2 sided-polygons; does not mean that removing 2 sides from a rectangle produces a digon or a 2-sided polygon.

@kaywisehero @Perrid13 en.wikipedia.org/wiki/Digon

I used to think there are no stupid questions. But now I increasingly realize that they indeed exist. They’re the asinine, ambiguous “gotchas” you use to claim AI is dumb because its perfectly reasonable answer didn’t match your own particular answer.

@slopwareindy @koltregaskes By existing benchmarks. While it is true that Sonnet and Opus are not in the same class, AI companies do try to make it seem like the difference is not huge in their Sonnet/mini classes based on benchmarks. Swe bench verified Sonnet 4.6 79.6% vs Opus 4.6 80.8%

@koltregaskes By whom? sonnet is not even the same class of frontier level model as 5.5.

Many developers have suspected for months that GPT-5.5 outperforms Claude Sonnet for coding. But SWE-Bench reported near-parity, and it made people question what they’d been seeing in practice. DeepSWE aligns more closely with that day-to-day experience: GPT-5.5 scores 70% versus Claude Sonnet at 32%. That difference is substantial. DeepSWE focuses on what tends to matter in real workflows: whether an agent can take a short behavioral prompt, locate the correct area of the codebase, and implement the change cleanly - without needing you to enumerate files, modules, and functions. SWE-Bench often fails to capture that, due to dataset contamination and weaker verification. deepswe.datacurve.ai/blog

@josephathomas @theo Though greater than 24 hours early cutoff is properly too much for timezone difference to fully explain - with very rare exceptions which should not apply since neither are on either extreme ends of UTC offset timezones.

@theo I was under the impression that he used GMT or UTC because I’ve noticed my resets happen around 6 PM central

My Claude Code sub expires tomorrow. I barely use it, but I still had it installed on my Windows PC so I used it to debug some crashing earlier. They hard cut me off over 24 hours early.

@laurie_guilbeau @NepsisVT @todayyearsold But since the question is only about number of sides left, does it still need to be a shape? For example, we would say that a flat piece of paper has 2 sides - front and back - even if it isn't a particular shape - since it represents a plane.

@NepsisVT @todayyearsold I thought the answer was zero because if two sides are removed from a rectangle you no longer have a shape since a shape must be enclosed

Am I stupid or is that literally correct

@weswinder To be fair, asking the model to self-identify is already the wrong approach. How many more variations of Gemini 3 Pro identifying itself as Gemini 1.5 Pro do we need? Would be better if he just pulled up an official source for model list and checked.

lol this dude just blocked me for telling him gpt-5.5 doesn't have a codex model variant?? he was (understandably) confused because codex (the app) identified itself as "codex" in chat openai has done irreversible damage by using the term "codex" for a million different things

@christoaivalis @lior_eth To be clear, I'm only talking about optimisation from what is sufficient/typical. So we aren't comparing an athlete with PED but given zero food, vs an athlete without PEDs but given optimal food. Obviously, zero food/rest/training are potential breaking on their own.

@christoaivalis @lior_eth For a non-athlete, it might be hard to gauge what feels subjective when we toss around labels like "small" and "huge" edge. But I'd propose that an edge can be considered huge if its effect is greater than the sum of what is obtainable via food, rest, training optimisation.

What the Enhanced Games have shown in real time: 1) everyone is already enhanced. Thor couldn’t even improve his deadlift by >1%. Either steroids are useless or he was already juiced to the gills. 2) enhancement without star athletes is kind of sad? “This 44 year old swimmer almost beat his personal best from 10 years ago!” is not a great look lol. 3) people are drawn to sports, particularly when the idea of performance & enhancement is not taboo. For the first event, I think it’s going well. Will be cool to see this thing grow.