I do not believe those backdoors are there because of some developer's mistake. But its actually placed there as backdoor to all servers.
Hard to believe...
Oh boy … PHP sent clear text data when you used uppercased S in httpS:// in a protocol handler 🙈🙈
file_get_contents('HTTPS://…') skipped TLS: tokens, POST bodies and FTPS logins went out unencrypted!
Fixed today in 8.2.36, 8.3.37, 8.4.28, 8.5.13 & 8.6.0.
@BullTheoryio Someone probably get really sad but i will not feel bad about those because they are greedy, tech is way more volatile and risky area.
Next year thing could get 5x efficient, if so, what all those old hardwares goes?
That could be DS 4.1 Pro.
@deepseek_ai just waiting right moment to release it maybe to trigger crash of NVDA to 94 levels.
US AI research/companies are still expensive IMO. If deepseek come with most advanced one it could be the one who trigger the crash
@OpenDesignHQ@Tuola0205 Is that just a skill or includes all tools as dsh plugin? If it is i would like to add into my harness. or have --profile design
@FluidVoiceApp@huggingface Give us StS one which run under 4GB vram with good tool call and less than 300ms.
The used model should not need to know how to code, how to paint how to translate... It just need to know how to talk generally and good tool call.
I'm get bored to switching tabs/sessions...
Why pp saying what not to be done? You know its easy.
Here what should be done;
Keep storing IDs (if you don't engine will do to optimise via assigning primary key)
Keep store EVM addresses. And have UUID of course to let each user access multiple addresses.
@svpino@Checkmarx It is expected, because people writing code manually are slower than those who let AI the write the code.
If each 10 code includes 1 bug, when you generate 1000 code thanks to AI you got 100 bug, in the other hand old-school dev just got 10 bug because he produce 10x less code.
@HigurashiAI Its good but couldn't you figure out conversation focused smaller model than Qwen3.5-4B with good tool call. So it could call bigger models if something else needed or confidence drops regarding topic what we talking.
Okay I was using all those manually but have to figure out more autonomous consumption of all those APIs.
Otherwise I can not consume those credits if i were manually prompt. Even I manage multiple projects at the same time with sub agents.
Recently I'm seeing inteligence jump
@thdxr Funny... there are maybe 100 sandbox company, 10+ open source sandbox software.
Providing inference at large scale is harder than wrapping AWS EC2 with your sandbox software and trying to scam pp by asking 10x money.