swordgardctf

Pentesting, Software Engineering, Machine Learning, Agent whispering, competing and more. Opinions are my own.

[swordgardctf @swordgardctf]

@HackingLZ Perhaps he did and that's what led to the HF incident ;)

[swordgardctf @swordgardctf]

@yacineMTB Opus at work, sol for personal projects. Sol is still my favorite by a mile except for UI design (rarely have this use case) where getting views from different LLMs is better, but sol still implements. Sol is just so good and cheap as an implementer.

[swordgardctf @swordgardctf]

@yacineMTB Honestly I'm surprised more people haven't come to this conclusion already on their own lol

[swordgardctf @swordgardctf]

@PhiloGroves same

[swordgardctf @swordgardctf]

100%

[Philo Groves @PhiloGroves]

I don’t know why we need browser extensions and browser-use integration when programmatic CDP is more powerful, and at this point I’m afraid to ask

[swordgardctf @swordgardctf]

@yacineMTB @SebAaltonen Outcomes needs to be consistently measured. Ideally not just from a performance but other metrics. Build hooks in separate code-bases that it cant modify to catch the model drifting/cheating.

[swordgardctf @swordgardctf]

@HackingLZ This is bound to happen.

[Ahmad @TheAhmadOsman]

Harness Engineering is the new Software Engineering

[kache @yacineMTB]

Pure reinforcement learning is what really scares me right now. All this language model stuff is cool but reinforcement learning working, from scratch. It's going to change the world

[vik @vikhyatk]

if i had AGI, i wouldn't tell anyone. but there would be signs (my uptime would exceed one 9)

[swordgardctf @swordgardctf]

My feeling today as well.

[Awni Hannun @awnihannun]

I have the complete picture

This changes the picture entirely

[swordgardctf @swordgardctf]

@IceSolst @SwiftSecur1 Don't know if it's rage bait so i'll bite. This list doesn't align with any specific engagement-types such as for netpen (either external or internal), web, mobile, RE, or any of the more specialized types of testing from my experience. Happy to discuss in good faith though.

[swordgardctf @swordgardctf]

@0xTib3rius It's just slightly less subsidized, still very subsidized. The cost of inference for large models is just absurdly high.

[swordgardctf @swordgardctf]

@0xTib3rius Tokens weren't going to be subsidized/free forever. VCs want their 1000x ROI sooner or later.

[swordgardctf @swordgardctf]

@HackingLZ Security companies right now: Do as I say, not as I do.

[swordgardctf @swordgardctf]

@0xTib3rius You probably forgot to turn on karmic dice IRL.

[swordgardctf @swordgardctf]

@HackingLZ @techspence It's almost as if bad offsec engagements without context provides negative value to companies.

[swordgardctf @swordgardctf]

@__tinygrad__ Any way to integrate with MLX?

[swordgardctf @swordgardctf]

@curi0usJack Well technically it's not a security control, its a targetting mechanism for arbitrary policy on AD. I would say this is more about AD being prevalent for both user and computer management than GPO in itself.