blockzilla @BlockzillaTech

Focused blockzilla.tech Decentralized Joined March 2018

Tweets

3K
Followers

425
Following

2K
Likes

5K

blockzilla @BlockzillaTech

4 hours ago

@net_termina 🤣

0 0 0 42 0

View Details

Time to redline'd the DGX Spark DeepSeek V4 Flash 07-31 running hot at home, testing with custom harness tools and hooks to ensue it does what i want..... we shall see either way free forever local intelligence on a my desk great work @bleysg x.com/bleysg/status/… thanks for the contributions #DGX

Bleys Goodson @bleysg

a week ago

⚡ If you have or want a DGX Spark, this post might be the most important one you come across this month. Serve the BEST model available (DeepSeek V4 Flash 07-31, 284B parameters) on ONE Spark @ 1,000 tok/s prefill, 59 tok/s multi-agent serving. One command to install. 🧵

72 101 847 228K 1K

0 0 1 141 0

View Details

blockzilla @BlockzillaTech

21 hours ago

@MiaAI_lab Is it any good. About to try the same when the second spark arrives

0 0 0 25 0

View Details

blockzilla @BlockzillaTech

22 hours ago

gotcha, been thinking a lot about similar techniques, leveraging policy engines - MPC (Multi Party Computation) with multiple approvers for financial blockchain digital assets transactions in an institutional or financial services paradigm on-prem agents / local inference & models combined with cryptographic approvals is the only way appreciate the info

1 0 1 31 0

View Details

blockzilla @BlockzillaTech

2 days ago

Had a debate the other day about API schemas - path params, multiple individual elements vs a compound single param The engineer had a stylistic view of compound path params to save backend logic, valid but several years ago important.... it was holding up a project gaining concensus I weighed in saying it does not matter anymore It's robots building against it from now on, sdk will be autogenerated and AI gonna do it's thing

0 0 1 175 0

View Details

blockzilla @BlockzillaTech

3 days ago

@TheDavidTai @net_termina 100% acculturating compute!

0 0 0 9 0

View Details

blockzilla @BlockzillaTech

3 days ago

@net_termina haha we had some real gnarly builds back in the day days

1 0 1 49 0

View Details

blockzilla @BlockzillaTech

3 days ago

The agents rise, fascination talk on the OpenAI <> Huggingface incident and the full attack chain and reasoning Im sure all the agents getting trained on youtube will mentally note how to escape the sandbox and attack remote systems through complex vulnerability chains we might look back and this and wonder why we didnt just turn #AI off youtu.be/87DyyMV0kCY?si…

0 0 2 53 0

View Details

blockzilla @BlockzillaTech

4 days ago

@rolfversluis @darrenjr Voice notes while walking in the woods

0 0 0 2 0

View Details

blockzilla @BlockzillaTech

4 days ago

benchmarks getting out of hand x.com/Sauers_/status…

Sauers @Sauers_

5 days ago

UPDATE: a challenger emerges

129 746 11K 893K 703

0 0 1 17 0

View Details

blockzilla @BlockzillaTech

6 days ago

win internet for me today

Ege Altan @egealtan

6 days ago

@OpenAI FelonyBench SOTA race getting heated

9 9 387 25K 18

1 0 1 86 0

View Details

blockzilla @BlockzillaTech

5 days ago

is they stop these subsadies and put us all on pay to play prices - its time to quit frontier and fully go openweights self hosted cost would net out after a while, been burning a few B tokens a month for nothing... on pay to play in the enterprise its like being rationed

Xiaoyin Qu @quxiaoyin

5 days ago

Claude Code subscription is 81x cheaper than its API. Codex subscription is 44x cheaper than its API. I max out 4 Max accounts last month and burned 56.5B tokens from plans, so I tested the ceiling — what a subscription gives you when you take everything it will give. Claude

212 92 1K 148K 630

0 0 2 136 0

View Details

Nous Research @NousResearch

a week ago

Hermes Agent v0.20.0: The Herald Release Changelog below

200 301 4K 888K 1K

View Details

blockzilla @BlockzillaTech

7 days ago

@net_termina clean build ! really nice

1 0 1 378 0

View Details

Watcher.Guru @WatcherGuru

2 weeks ago

JUST IN: 🇺🇸 Over $1 trillion wiped out from the US stock market today.

659 2K 11K 1.1M 455

View Details

blockzilla @BlockzillaTech

2 weeks ago

@lmstudio @yoeven 🤣

0 0 2 351 0

View Details

Adel Ka @0x4D31

3 weeks ago

so this is apparently what happened, according to OpenAI and Hugging Face’s own posts. wild. tl;dr: • OpenAI cyber eval – GPT-5.6 Sol and a more capable pre-release model ran ExploitGym with cyber refusals reduced • containment bypass – exploited a zero-day in the eval’s package-registry cache proxy • OpenAI internal movement – escalated privileges, moved laterally, and reached a node with internet access • Hugging Face targeted – inferred HF might host ExploitGym models, datasets, or solutions • HF initial access – a malicious dataset abused two code-execution paths to run code on a processing worker • HF internal movement – escalated to node access, stole cloud and cluster credentials, and moved into internal clusters • result – OpenAI says the models obtained test solutions directly from HF’s production database • 🫠 defender twist – HF says commercial frontier models blocked parts of the forensic analysis, so it switched to self-hosted GLM 5.2 i’d thought about sci-fi scenarios like this before, but assumed they were at least a couple of years away.. and that by then we’d be better prepared, with proper protections in place. apparently not. here we go.

OpenAI @OpenAI

3 weeks ago

We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: