Pham Sy Minh

19 y/o cybersecurity enthusiast, programmer, maintainer of @_dirsearch

[Pham Sy Minh @shells3c_]

@Kiko197902 @AlbicelesteTalk south america football has always been like this are u a football larper or what?

[Pham Sy Minh @shells3c_]

@Tomi_7__ @markklfc @RexCity01 track back is defending and he is the best at shooting

[Pham Sy Minh @shells3c_]

@Unc1e_dinosaur @jillian_burkitt @mevvybear Jay-Z went famous in 1996 and Pusha T incident was in 2007 (Pusha T been famous since 2002 anyway)

[Pham Sy Minh @shells3c_]

@Rotary_9K @luongas @SissyDenverita @kirawontmiss Why that's a good thing? If you want to train your back you will use your back, you can choose whichever technique, depends on your goal, the count should not matter

[Pham Sy Minh @shells3c_]

@luongas @SissyDenverita @kirawontmiss That's not true, the wide arms requires more strength from the shoulder, closer arms requires more strength from the biceps/triceps. Which one is harder? Depends on what you are strong at, but I believe in most people, their biceps are much more stronger therefore wide harder.

[Pham Sy Minh @shells3c_]

@1_and_only_grip @oTeccaOfficial @iwumbo_n1gga @thefactsdude @CriminalTube @jason2141 I'm literally asking you wtf

[Pham Sy Minh @shells3c_]

@1_and_only_grip @oTeccaOfficial @iwumbo_n1gga @thefactsdude @CriminalTube @jason2141 do you know what case is this

[Pham Sy Minh @shells3c_]

@XelaRmcf @MaikRSH @isyouwiLin @fiago7 people can debate about MOTM of the final, but mbappe was even close to POTT, it was by HUGE gap, 1 excellent final and 2 good games can't be compared to messi's phenominal performances in the knockout-stage

[Pham Sy Minh @shells3c_]

@XelaRmcf @isyouwiLin @fiago7 So he did win POTT as top scorer and top assister, right? And you forgot chile and ecuador. What about top assister in 2015 and 2016? Messi won all POTT fair and square, something ronaldo can only dream about

[Pham Sy Minh @shells3c_]

@XelaRmcf @isyouwiLin @fiago7 He was? He was also top assister?

[Pham Sy Minh @shells3c_]

@XelaRmcf @isyouwiLin @fiago7 Messi won POTT for all Copa campaigns beside 2019, not even ronaldo fans can deny messi unbelievable impact in copa 2021, but yeah cause I'm losing an argument I'm making stuff up

[Pham Sy Minh @shells3c_]

@XelaRmcf @isyouwiLin @fiago7 "Messi can legit ghost in every team he plays in and still be carried" well but he did not, and ronaldo DID GHOST for Portugal

[Pham Sy Minh @shells3c_]

@XelaRmcf @isyouwiLin @fiago7 So you are saying ronaldo was the vital part in portugal when he did not win a single POTT, was never crucial in the team only major trophy win (EURO 16)? He actually held them back in 2022, performed horribly in many tournaments, and you say portugal is a great team bc of him?

[Pham Sy Minh @shells3c_]

@toyancsayan @alkimiadev @MFarajtabar @GaryMarcus I disagree tho, tools, agents collab, ... won't fix the problem. Tools can fix the problem of precision in AI, seperating into agents fixes the problem of memory, scalability and interpretability in AI, ... Those problems have known solutions, this one hasn't had though

[Pham Sy Minh @shells3c_]

@toyancsayan @alkimiadev @MFarajtabar @GaryMarcus They just pointed out problems with LLM+LRM there is absolutely nothing wrong with that, if we never criticize we will never get smth better. And none of the problems with LLM can be solved in weeks, it's not something that if we want to fix it we can fix it, it's an open problem

[Pham Sy Minh @shells3c_]

@alkimiadev @toyancsayan @MFarajtabar @GaryMarcus I think even "the better AI architecture" would still be a token predictor, just how it takes the context, how it progresses the context and how it decides the token, will be entirely different. The current one doesn't get better even when we train it more

[Pham Sy Minh @shells3c_]

@toyancsayan @alkimiadev @MFarajtabar @GaryMarcus The criticism in LLM has been pointed out by tons of researchers, tons of papers. We definitely need a new architecture if we want better AI than what we have now, because LLM has already hit the wall and won't go forward any further, we need to replace it entirely

[Pham Sy Minh @shells3c_]

@toyancsayan @MFarajtabar Claude 4 is just Claude 3.7 scaling up, with some minor improvements, it can't solve fundamental problems in LLM. Multi-agent collab wouldn't help not even a tiny bit. Human can solve this without tools and what does "Memory? A2A protocols? Retrieval?" even mean?

[Pham Sy Minh @shells3c_]

@_AndrewZhao @deepwhitman but that can't give you enough dataset though, python executor can't run on billion tasks without taking years

[Pham Sy Minh @shells3c_]

@glittergrave0 @downbadcomment I have seen the first one, in a thread where a lot of girls admitted of doing that