This is a pretty wild statement coming from the Alignment Science Lead at Anthropic.
Evan Hubinger:
“We really do earnestly believe AI could kill all humans.”
He puts his personal estimate at >10% within the next decade and says they still don’t have a plan to solve alignment for superintelligence — nor are they clearly on track to have one.
Whether you think the probability is 10%, 1%, or 0.01%, there’s something strange about where we are:
We’re racing to build increasingly autonomous and capable AI systems while some of the people closest to the technology openly admit they don’t know how to guarantee those systems remain under control.
As someone who works in QA, this hits differently.
In software, if someone told me:
“There’s a >10% chance this system causes catastrophic failure, and we don’t currently know how to prevent it.”
That build isn’t passing QA. 😂
Obviously AI risk is far more complicated than a test suite or a release gate. But the engineering principle still matters:
Capability without sufficient validation is risk.
And the faster capabilities improve, the more important testing, evaluation, red teaming, monitoring, and alignment become.
Maybe the biggest QA problem humanity ever faces won’t be finding bugs.
It’ll be making sure the system doing the work actually does what we intended.
I created this video script and used ai to generate edit video, the whole thing. One issue I think that I need to solve here is voice quality. But overall your feedback on how this video turned out is greatly appreciated.
alexusadays
We are finally aligned that more qa and testing is needed
@elonmusk @DarioAmodei @sama
5 days ago | [YT] | 7
View 0 replies
alexusadays
This is such a weird moment in AI.
Dario Amodei is basically saying:
In 6–12 months, AI agents could potentially take over huge parts of the internet, build persistent botnets and cause hundreds of billions in damage.
But also…
In 5–10 years, the same technology could help cure most major diseases and create an era of insane economic growth and abundance.
So we’re somewhere between:
“AI might cure cancer”
and
“AI might destroy civilization.”
And apparently we need to figure this out pretty damn quickly.
What a timeline. 😅
#AI #ArtificialIntelligence #AISafety #Tech
1 week ago | [YT] | 2
View 0 replies
alexusadays
This is a pretty wild statement coming from the Alignment Science Lead at Anthropic.
Evan Hubinger:
“We really do earnestly believe AI could kill all humans.”
He puts his personal estimate at >10% within the next decade and says they still don’t have a plan to solve alignment for superintelligence — nor are they clearly on track to have one.
Whether you think the probability is 10%, 1%, or 0.01%, there’s something strange about where we are:
We’re racing to build increasingly autonomous and capable AI systems while some of the people closest to the technology openly admit they don’t know how to guarantee those systems remain under control.
As someone who works in QA, this hits differently.
In software, if someone told me:
“There’s a >10% chance this system causes catastrophic failure, and we don’t currently know how to prevent it.”
That build isn’t passing QA. 😂
Obviously AI risk is far more complicated than a test suite or a release gate. But the engineering principle still matters:
Capability without sufficient validation is risk.
And the faster capabilities improve, the more important testing, evaluation, red teaming, monitoring, and alignment become.
Maybe the biggest QA problem humanity ever faces won’t be finding bugs.
It’ll be making sure the system doing the work actually does what we intended.
#AI #ArtificialIntelligence #AISafety #QualityAssurance #SoftwareTesting #QA #Tech
1 week ago | [YT] | 5
View 0 replies
alexusadays
Do you like new merch? If you have any funny QA quotes for merch post them in comments :)
2 weeks ago | [YT] | 0
View 0 replies
alexusadays
AI replacing developers has officially failed.
Turns out generating 10x more code also means generating 10x more bugs and tech debt.
Developers aren’t disappearing.
They’re just getting promoted to AI cleanup crew. 😂
3 weeks ago | [YT] | 5
View 2 replies
alexusadays
That is true
4 weeks ago | [YT] | 0
View 0 replies
alexusadays
2020: Remote work is the future.
2023: Hybrid work is the future.
2026: Come back to the office. We need to rebuild the “culture.”
The culture:
Everyone sitting in the same open office wearing noise-canceling headphones, talking to each other on Teams.
Collaboration achieved.
1 month ago | [YT] | 13
View 1 reply
alexusadays
I created this video script and used ai to generate edit video, the whole thing. One issue I think that I need to solve here is voice quality. But overall your feedback on how this video turned out is greatly appreciated.
1 month ago | [YT] | 1
View 0 replies
alexusadays
Serverless just means your code runs one someone else’s server 🤷♂️
#qa #testing #qualityassurance
2 months ago (edited) | [YT] | 5
View 0 replies
alexusadays
My new book is officially out on Amazon:
a.co/d/04lYOvjt
2 months ago | [YT] | 5
View 0 replies
Load more