Archwizard César

The company has concluded its obligations and has no further transmissions to send. The establishment continues to function, its service to the community unwavering, despite the serpent's negative influence.

till we meet again.. i love frieren beyond words, what a beautiful end to season 2 ♡ #frieren #indie_anime

@StarCraft I remember spending a summer playing UMS custom maps in Brood War.

March 31, 1998...what is your favorite memory from StarCraft's first 28 years? 🎉

resource allocation requires visible failure

@Niftski #FreeNiftski ✊ reverse the vote

This video proves beyond a reasonable doubt that current Any% WR holder averge11 conspired with certain SMB1 leaderboard moderators and others to rig the vote to ban input swapping for the purpose of sabotaging my Any% TAS tie progress. youtu.be/e9Kx0kw6Iq0

@fchollet Francois would it be possible to update the games to show the number of steps it took you to win or lose?

ARC-AGI-3 is out now! We've designed the benchmark to evaluate agentic intelligence via interactive reasoning environments. Beating ARC-AGI-3 will be achieved when an AI system matches or exceeds human-level action efficiency on all environments, upon seeing them for the first time. We've done extensive human testing that shows 100% of these environments are solvable by humans, upon first contact, with no prior training and no instructions. Meanwhile, all frontier AI reasoning models do under 1% at this time.

painting what do you see?

@RiotPhroxzon I need you and your team to write a technical book or a series of articles about this matchmaking saga, because I've loved following along and learning about the topic and how you've approached the problem. Please consider it!!!

Apex Ranked Followup Thank you for all your feedback about the changes last week. I wanted to give some explanations on what we're seeing and why it is how it is right now; I’m not trying to change anyone’s mind, but I hope being transparent will lead to more constructive conversations with everyone Feedback we heard 1. The ladder has less meaningful breakpoints to strive for, now that the LP gaps between Master, Grandmaster, Challenger and Rank1 are really wide in a few regions (NA, EUW, EUN mainly). The gaps between tiers can feel exhaustingly large with low feedback and satisfaction on the journey from say low masters to high masters 2. It also makes comparison to previous season benchmarks lose meaning (1k LP, 2k LP, etc.) 3. Many are calling for an Apex Ranked reset; I'd love to know more about exactly what you mean by this (more below) 4. The ladder already felt grindy, like you had to play a lot of games to get to the next tier, and now it feels even more so 5. The top players getting +30/-10 even if their MMR is high feels unfair if a new or lower account can't do that; there are feelings of “how can I catch that” 6. Depending on which patch someone played, with the same winrates, their LP outcomes can be quite different, which is frustrating On who is getting +30/-10 and who is +/-20 - There have been a lot of discussions around who is getting good gains and who is not - We agree it feels unfair right now for the top of the ladder to be getting +30 while others are getting +20; I just wanted to explain why this is - This is because the weeks many players spent eating +10/-30 from the soft cap is being repaid; essentially for every game that a player played a +10/-30LP game, they will get paid back with +30/-10LP ones and this will grow the top of the ladder (similar to how the max LP on the ladder grows early in the season) - Once the ladder stabilizes, 95+% of the ladder (including the top of the ladder and including masters entrants at the bottom) is intended to get +/-20 - This means the only way to climb the ladder is to have a >50% winrate - If you have a 50% winrate over a long period, then you’re probably in the right skill level and are not in a climbing state - I also want to state this very clearly as a response to folks saying they should roll a fresh account to fight the Challenger LP gains. There is no advantage to running a fresh account up the ladder to try and hit an Apex rank, it will always be better to start with a pre-existing Apex account - I know it didn’t work like this in Seasons past, but it does now (and has for the past season or two) and this is to further disincentivize smurfing, something many players on the ladder had mentioned as a pain point - The only way to climb from this state is to improve skill level - I can guarantee that a Challenger player will be able to climb just fine with +/-20 given enough games, because they will have a very high winrate through Master and Grandmaster, but this leads me to my next point On ladder grindiness - We hear your concerns on needing to play too many games to climb up the ladder - It is true that older accounts that played their accounts up to challenger will be advantaged in Season reset races with the way we currently do soft rank resets - We do this because we want to make camping spots less effective of a strategy, dissuade smurfing, and encourage people to play on their main accounts - If the legacy accounts are not advantaged, there is no blocker to just running many fresh accounts through the Ranked ladder to hit Challenger; I think most players would agree that would be a worse experience - Secondly, as soon as Challenger players run into negative LP gains, many will stop playing on their Challenger accounts and move to smurfing, which is bad for match quality and queue times as well - We believe the high LP values are a better alternative to negative LP gains, but they are both not ideal - Additionally, we have daily game requirements and cap the max LP gains at 30 so that players don't camp on their spots without playing so that others have more opportunities to overtake them - In a world where a new Masters account has a 75% winrate through the Apex ladder (ie. is probably a top 10 player), that is 300 games to get to 3000LP from 0LP - If you are starting from a legacy account, it will be significantly less games than this - We don't believe it should be possible to be able to get to Rank 1 from a fresh account in less than 2-300 games. That makes smurfing, running up multiple accounts and maintaining them too attractive of an option - For one of the premier competitive games, we don't believe it is too much to ask for a player to play 1-2 games a day (between 3-700 games a year for the highest skill players in the game). Genre expectations to reach the top in many other games (including other MOBA’s) can be orders of magnitude higher than this and often require full time grinding - On the flip side, we acknowledge that there's a sweet spot on how much a player needs to play to not perceive it as too grindy, many people have to study, have jobs, etc and so it needs to be achievable for them too - We want to balance all these considerations; reducing incentives to smurf, how grindy it feels to achieve/maintain a rank and how legacy accounts are treated On why the LP is so high - I saw a comment asking whether the gap between Iron and Master (2800LP) is really equivalent to the gap between Master and Rank1 Challenger (2, 3, 4000LP) - In some regions, the answer to that is yes, in others, it's not quite as large, but still close - Players have gotten significantly better each year, especially with how often the top players are boot camping, taking a shot at Pro and learning from it, and pushing each other to get better - This is one of the reasons why the LP gaps between tiers are so high and the existence of the soft caps in previous years ended up suppressing the observed top LP's by some amount, so the gap looked lower than it actually was - Factually, there is a huge gap in skill between Master and Grandmaster and again from Grandmaster to Challenger so amount of points between them has to be reasonably large - This is a very common pattern in long running games, for example in Chess, Magnus Carlsen vs any random Grandmaster has close to a 90% chance to win - As League goes on, the gap between Rank1 (say Showmaker) and Master 0 LP is going to continue to widen; there are so many things you can do to influence the team in small but meaningful ways that aren’t super noticeable individually but have a huge impact over the course of a game, like pinging, shotcalling, soaking pressure, getting vision, etc. - But there's a fine balance here, we can agree that progression between tiers can feel daunting in the current tuning and there is a lacking sense of progression. This is why we’re considering adding additional tiers to break this up and create more “checkpoints” On Matchmaking Quality - There are some expectations of being able to have full challenger lobbies, all duos balanced, all role parity (on-role vs off-role), low queue times, all equal LP, remove autofill at all times of day - I want to set an expectation that this is not possible with only 300 Challengers and 700 Grandmasters in many regions - Players need to be autofilled, especially at the top of the ladder for us to make queue times reasonable, but we can at least try to make those autofills balanced in role - If a game is unbalanced in one of the axes above, we try to balance it out in another axis, but we are sometimes going to have to grab some Masters players to fill Challengers lobbies (hopefully not during peak time) - Especially with the new role parity algorithm, we believe we are making very fair games (close to 50% chance to win) in >90% of situations, with close LP between teams, duo balance, role parity - We believe the new algorithm is already significantly better than the old one, even though there may be some rose tinted glasses about how much better Matchmaking was before, which we don't agree with. We are still improving it Why do Challengers get +30LP, even when there are 200 LP masters in the game - LP gains are given based on how fair the match is and mentioned above, over 90% of matches have 50% chance to win - The reason why the Challengers are getting +30 for these games is because of the repayment of debt in the points above; this will resolve itself soon and the players will quickly go back to +-20 - If the match itself is 50% chance for either team to win, then the performance of the various people in the game is already baked into the LP gains (ie. the 200 LP master is expected to play worse, the Challenger is expected to play better) - There will usually be something offsetting this LP imbalance (whether it’s an extra duo on one team, someone playing secondary instead of autofill, etc.) - As I mentioned above, we believe >90% of our games are fair; it can be hard to guarantee fairness in off-peak and/or in small regions Other things we're thinking about (nothing confirmed) - [Agree] Many players are calling for better feelings of progression and progress in these tiers - [Agree] Reductions of grindiness (eg. more decay game banks, increasing max LP gain past +30LP, lowering distance between tiers, adding new tiers) - [Agree] Better reasons to maintain and play on Challenger accounts, rather than Smurf - [Uncertain] Adding more Grandmaster/Challenger slots to regions that have high numbers of players (which would bring the points between tiers down) - [Uncertain] Reducing how much advantage players get on their legacy accounts from start of Season (eg. capping at +25LP at max, instead of +30LP), but this will also further incentivize smurfing and increase feelings of grindiness - [Agree] Lower the amount of resetting at start of Seasons (eg. maybe start the Season at Master 0LP) More on Apex resets - To get a better understanding of what y'all mean by ladder reset, some possibilities are detailed below, - Not committing to any particular action or if we would even do any of these, but we want to better understand your intent when some of you ask for a reset. We definitely are far out from talking about a “when” at this point - If we went forward with any of these we would only reset a few regions as the vast majority of regions have had a normal season - We will be doing some research in the affected regions to help inform a call one way or the other - We would only consider a reset if we are confident it would result in an improved overall experience Option 1: Hard Reset - Early matchmaking will be a cluster****. There would be no memory of previous season ranks in Matchmaking - This means you could have 5 exChallenger vs 5 Master peakers, and that would be considered a fair game in the system - Even if a player is Challenger, they might have a team of Masters and be unable to carry hard enough vs a pretty stacked team on the other side, making the climbing process feel very RNG - This matchmaking quality would go on for months as the ladder sorts itself out, which would contribute to a negative experience for a good amount of players - Early season this year was a bit of a taste of this as we did a bit of a harder reset, and matchmaking quality would be significantly worse than that. This would be the most extreme option - We still don't believe this is a good idea, but if y'all are still wanting to push for it given this context, then the team can continue to discuss it Option 2: Softer Resets - Soft Reset would keep some semblance of normalcy in matchmaking, but the best players will be rewarded for being high on the ladder with better position on the starting blocks so to speak. Previous challengers would get increased gains (+30/-10) well into their climbs - The softest option would be everyone keeping their relative positions in the ladder but would need to maintain their current winrate to prove they belong there and reach their previous LP value

It's time for Europe to wake up, come on G2!

G2 DOING THE IMPOSSIBLE WHAT IS HAPPENING #LeagueOfLegends #LastStand

@teortaxesTex The same thing is often done in the US, where some executive or the company itself takes credit for a major discovery instead of highlighting the actual key figures or team responsible.

It's so ridiculous when people say "Chinese scientists", "Chinese engineers" or "China" and just post Xi + some amazing success. This guy isn't personally inventing shit. Seriously tired of seeing his mug in non-political contexts

@mimi10v3 Dune?

what else belongs in my tpot book collection?

@Blizzard_Ent @CallMeTasteless @Artosis WE ARE SO BACK LFG

The Blizzard Classic Cup needed captains… so we called in the legends. Welcome back, @CallMeTasteless and @Artosis. The iconic duo returns for legendary calls and unforgettable moments at BlizzCon 2026. More details on the Blizzard Classic Cup coming soon 👀

@SOOP_EN the ability to reset password or register a new account seems to be broken for me, please help I want to watch @Artosis I opened a support request too

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

I tried vibe coding a tiny world simulation where LLMs pilot the agents. One of the features I added was the ability to pray to the creator of the world (i.e. me), and this was the first prayer I received from the agents!

@clehene @ComputerHistory @kevinlynch Thank you for the response!

@CesarAndreu @ComputerHistory @kevinlynch is a good starting point for Flash history and he's been also influential to the fate of the project I mentioned.

@clehene hi I saw this post on HN (news.ycombinator.com/item?id=472551…) have you ever considered going to a technical podcast to share some of these stories? it sounds pretty interesting and I feel like a lot of the history around Flash is underexplored relative to its massive impaact

@BlizzHeroes WE ARE SO BACK POG

⚔️The Blizzard Classic Cup featuring StarCraft, StarCraft II, Warcraft III, and Heroes of the Storm is coming to BlizzCon 2026! #BlizzardClassicCup #ClassicCup26 #LegendsReturn #BlizzCon

I'm most interested in learning what lessons some of these older legacy software toolkits have to teach modern development which might not have been properly communicated previously. Love the podcast, thanks for everything!!!!

.@wookash_podcast Stellar interview with Chuck Jazdzewski, I love the history lesson on UI toolkit evolution! Any chance you could try to get some of the engineers that helped create Macromedia Flash or Apple's Cocoa as guests?