Samuel Cardillo @CardilloSamuel

Leveraging technologies to execute visions | Former(?) CTO @RTFKT @Nike | Ex-@Google | CEO @sbreakintl 🛰 | Opinions are my own Free world Joined July 2011

Tweets

30K
Followers

43K
Following

2K
Likes

30K

Samuel Cardillo @CardilloSamuel

42 minutes ago

own your inference.

AshutoshShrivastava @ai_for_success

3 hours ago

It's so over, folks... First Fable 5. Now GPT-5.6. The Trump administration has reportedly asked OpenAI to stagger the release of GPT-5.6 over security concerns. Sam Altman reportedly told staff that access to GPT-5.6 will be approved on a customer-by-customer basis. At this

41 17 271 40K 64

1 0 2 383 0

View Details

Samuel Cardillo @CardilloSamuel

2 hours ago

@25_uzi @NVIDIAAI then for 6k usd you can!

1 0 0 17 0

View Details

Samuel Cardillo @CardilloSamuel

2 days ago

just had a call with yet another company switching to local ai. they were using Qwen 3.6 35b served through my spark so they could test their pipelines through it and see if its a fit and what i heard in that call is pretty crazy and i feel it is something that most anti-local ai misunderstand. what the customer told me is literally that they kept comparing the results between Claude, ChatGPT and the locally served Qwen 3.6 and largely preferred the results they were getting via Qwen - insane! the plan is now to deploy a DGX Spark in their office for the MVP + a 6000 pro blackwell-enabled machine for training & later on they're looking to basically buy a DGX Spark per customer they have so they can serve each customers an optimized finetune. this is the future. literally. and once again, this is the worst it'll ever be.

176 154 2K 217K 1K

View Details

Saarth Shah @saarth_

6 hours ago

We built an AI agent that traces child predators. A 41 year old man has been commenting on teenage girls' posts for 5 years. Our agents traced him from his Instagram handle. This kind of predatory behavior is happening at scale right now across Instagram, Discord, TikTok, Reddit, and other platforms. This isn't just for predators. Scammers, fraud rings, ban evaders, and counterfeit sellers can now be identified from a single identifier. If your T&S team can flag them, we can find them. Platforms have the content signals. What they can't see is who is behind the account and whether they've done this before. Our agents can.

122 115 1K 133K 329

View Details

Samuel Cardillo @CardilloSamuel

3 hours ago

@25_uzi @NVIDIAAI for 4k usd you can!

1 0 0 23 0

View Details

Samuel Cardillo @CardilloSamuel

5 hours ago

@nasdaily @sellwithnas This is easily one of the smartest extension for your company. You're absolutely killing it. Insane visionary! Absolute respect from me, men!

0 0 2 215 1

View Details

Samuel Cardillo @CardilloSamuel

8 hours ago

@nikxbt @0xSero wtf you talking about, nothing in the tweet says it took 28 minutes anywhere

0 0 0 22 0

View Details

Samuel Cardillo @CardilloSamuel

11 hours ago

just did the same test with @0xSero REAP version he recommended me (huggingface.co/0xSero/GLM-5.2…) Q4_K_XL, on 2* 6000 pro blackwell * 250 GB RAM, squeezing ~17.4 tokens per seconds (instead of 14) but it took 107 minutes to complete. and that comes with a very interesting conclusion: yes, the optimized REAP version might be outputting faster but, because of the tradeoff on intelligence means that i do end up loosing time than when using the UD-Q4_K_XL GLM 5.2 from @UnslothAI . so, if i want something developed "fast", i could use the REAP optimized version but if i want to make it better, i'll use the Unsloth version, with even a possibility, depending on the challenge within the code that needs to be written, to end up in the same kind of results i just had in this test.

Samuel Cardillo @CardilloSamuel

a day ago

16 13 205 29K 126

3 1 19 3K 7

View Details

Samuel Cardillo @CardilloSamuel

9 hours ago

@wbetiago tell me you're joking

2 0 5 424 0

View Details

Samuel Cardillo @CardilloSamuel

9 hours ago

@Renegade_Rocks the answer is somewhere in the many replies

0 0 0 139 0

View Details

Samuel Cardillo @CardilloSamuel

9 hours ago

@NewVaneckIntern tell me you have no f understanding of what you're talking about without telling me you have no f understanding what you're talking about

1 0 0 63 0

View Details

Samuel Cardillo @CardilloSamuel

12 hours ago

@b2kdaman 😂😂😂

0 0 1 36 0

View Details

Samuel Cardillo @CardilloSamuel

16 hours ago

one of the best thing happening to me before a us trip is the pre-timezone adaption that i do. i slowly shift my sleep schedule entirely so i end up being awake at night and sleepy during the day in dubai so when i land, i am not a dead corpse. why its amazing? absolute productivity by working at night without any distraction.

7 0 12 2K 0

View Details

Samuel Cardillo @CardilloSamuel

15 hours ago

@NicoGrassetto not even! because i cna't play without my 3 mates and they never there :v

1 0 1 62 0

View Details

Samuel Cardillo @CardilloSamuel

15 hours ago

i guess now you know where i am traveling next

0 0 3 693 0

View Details

Samuel Cardillo @CardilloSamuel

16 hours ago

@Elinoam_Shefi @amjadt25 is Israel part of the GCC ? no. stop confusing shit.

1 0 0 76 0

View Details

Samuel Cardillo @CardilloSamuel

16 hours ago

@codechad @AlexanderTw33ts man of culture

1 0 1 45 0

View Details

Samuel Cardillo @CardilloSamuel

16 hours ago

@charleshlau 🤝

0 0 4 239 0

View Details

Samuel Cardillo @CardilloSamuel

17 hours ago

@andrew_Tas did it got solved at some point?

1 0 1 167 0

View Details

Samuel Cardillo @CardilloSamuel

24 hours ago

as expected - and it is entirely normal - qwen 3.6 35b a3b completely failed at it. after roughly an hour at ~80 tps (served via my spark) it did a wrong tool call and killed itself. the approach of qwen somehow was doing a lot of reverse engineer through javascript (?) so it wrote a bunch of .js which honestly don't amount to much. interesting how, GLM 5.2 at ~14 tps just took a little bit longer to succeed.

Samuel Cardillo @CardilloSamuel

a day ago

16 13 205 29K 126

10 1 36 7K 11

View Details

Samuel Cardillo @CardilloSamuel

17 hours ago

@0xSero @samelldev @opencode i'm running the Unsloth GGUF UD-Q4_K_XL, loving it so far! i'll try out yours, seems interesting! i usually tend to stay away from REAP because i always have bad surprises (especially with looping and broken logic), will give feedback!

1 0 1 138 0

View Details

Samuel Cardillo @CardilloSamuel

a day ago

i've downloaded a crackme challenge (crackmes.one/crackme/6a3ab0…) and gave it to my local instance of GLM 5.2 running on 2* 6000 pro blackwell and ~300 GB RAM. used @opencode has the harness without any access to debuggers or MCPs. result: succeeded in 78mins (~14 tokens per second) without access to any debugger, it checked if the machine had python installed and started to dump memory addresses, analyze them, do manipulations, tests, ... and followed the instructions, not patching the crackme. this is impressive. honestly. i'll try with qwen 3.6 35b a3b next.