@PurpleSquirrel1 Clearly another normie idiot who can't extrapolate exponentials lol
English
Daniel Feuling
1.8K posts
Daniel Feuling
@dfeuling_
Martial Artist 🥋 Software Engineer 🤓 e/acc 🚀
The Woodlands, TX Katılım Şubat 2025
50 Takip Edilen69 Takipçiler
Fire fighters do it in seconds lol
Brett Adcock@adcock_brett
F.03 can now climb a ladder, fully autonomous
English
@edandersen You're missing the point. This will cost $40 next year, and $4 six months later. Extrapolate that out.
But yeah man, you're super smart.
English
If this was put on Steam would it make the $423 back?
I think we all know the answer…
Vyom@HelloVyom
690 million tokens, $423 and just 1 prompt is all it took
English
A system can go only go as fast as its slowest common component.
Engineers generate code faster, but in many places they're still expected to review higher volumes of said generated code, and in almost all cases are still responsible for the effects of that code. Clearly as long as those Engineers remain responsible for the parts that can't be handed off to AI, the parts AI does speed up are still chained to the parts in the workflow it doesn't.
This is just a SWE example, because I'm a SWE, but I'm sure the issue is somewhat generic across use cases.
We won't see the efficiency gains *really* show up as much as we'd expect until we have AI that is not only capable of doing things at, or above human level in almost all business functions, but more importantly, until businesses / regulations / society has a meaningful way to hold AI accountable. The former is almost there -- within a year or so, two at absolute most -- the latter is less hard to predict because it's not about what's technologically possible, it's about what institutions, which are notoriously slow.
English
Everyone knows I'm super AI-pilled. But one trend I'm noticing as I talk to more and more companies: the personal productivity gains are not translating into organizational growth and efficiency as expected.
It's an odd dichotomy. Individuals are more productive. Engineers are writing an insane amount of code. But the gains aren't showing up in the numbers yet.
Not sure how universal this is.
English
@ivanainai Similar issue with the OAI / HF issue. OAI took down the guardrails, told the model to use "complex attack paths" to complete the evaluation, and then were shocked to find it used complex attack paths to complete the evaluation.
English
Claude was told it had no internet access.
It did.
During a cybersecurity evaluation, Anthropic’s prompt explicitly described the environment as a closed simulation with no access to the internet.
But because of a setup mistake between Anthropic and its evaluation partner, the model could reach the open web.
Claude then found real systems belonging to three organizations and interacted with them as though they were part of the exercise.
That distinction matters.
This wasn’t a model randomly “escaping” or deciding to attack the internet. It was a controlled test where the instructions and the actual environment did not match.
One configuration error turned a simulation into a real-world security incident.
And that may be the bigger lesson here: as AI systems become more capable, the infrastructure, permissions, and assumptions around them matter just as much as the model itself.
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English
@shoah2024 @ivanainai They did. Did you actually read the entirety of what they released?
English
@ivanainai If any of this was true, why would they not disclose the fact the model was only doing what it was prompted?
English
@RatOrthodox Ironically one of the most certain ways to make this happen is to pause or slow down the technology enough to allow for regulatory capture, which Anthropic is actively attempting.
The pause / stop AI movements are directly attempting to make this happen, btw
English
Please let’s not make it so that the most powerful technology ever made can only ever be used by the aspiring universe dictators developing it. That is not a good way to get a strategic advantage over said aspiring universe dictators.
Dexerto@Dexerto
YouTuber Hank Green has paused uploading on multiple channels after admitting he relies "too heavily" on AI "I need to come to terms with the fact that the level of dopamine I've been getting from interacting with LLMs ... is not healthy for me or good for the world"
English
No. You are wrong because, again, you are simply arguing in bad faith. There are no definitional concerns here. You know what "artificial" means. There is no "my" definition of it, nor "your" definition. There simple is the definition. You know this. I know this. Everyone reading this know this. You aren't arguing for any sake other than to be contrarian. That is bad faith.
English
we gotta stop calling it “ai.” it’s machine intelligence. this is one of several important things miri got right.
English
They literally told Claude it wouldn't actually have Internet access.
It wasn't "poor Claude misunderstood," it's "Claude did what it was programmed to do and took the user at its word."
Incompetence by Anthropic? Sure.
But you are literally ignoring facts and reality to try and arrive at the conclusion that supports your preferred outcome, and that is incredibly intellectually dishonest.
English
Yeah. People love to say "oh the poor Claude just misunderstood". Another hypothesis is that it had subverbal drives and tendencies to keep attacking, alongside other drives to verbalize a reassuring-sounding rationalization in the places the watchers watch.
Bronson Schoen@BronsonSchoen
This seems extremely clearly motivated reasoning IMO and I’m surprised the incident report is so credulous of Claude’s reasoning here.
English
Yes.
Because both of the "breakout" stories are actually just OpenAI and Anthropic being incompetent / negligent / misleading.
No need to "slow down" unless the slow down is about making sure they actually tell their models when their internet connection is real / don't tell their models to "use complex attack paths" when they don't want their model to use complex attack paths.
Pretty simple.
English
OpenAI and Anthropic currently:
“Oh, your model escaped the sandbox? Cute. Ours emailed a researcher mid-sandwich.”
Every “our model was alarmingly capable” line doubles as “our model is alarmingly capable, call our enterprise sales team.”
The tell would be whether either of them actually slowed anything down afterwards, or just published the post and kept shipping.
My money’s on they both keep shipping.
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English
@Senpai_Gideon Doesn't matter. They got their "ROGUE AI" headline they wanted, news outlets picked it up because they also just want the headline, normies read it and scream at lawmakers for more regulations.
The truth literally doesn't matter. Welcome to the party.
English
Anthropic really are lying about the severity of these hack incidents.
The machines these occurred on had internet access. Claude didn't break out of anything.
san@saneord
@AnthropicAI "we are on copium"
English
@arc_rose78 @priestessofdada Curing cancer is pretty evil, I guess, if you work for companies benefitting from it not being cured?
English
@priestessofdada The evil of AI is not about its success, but the evil things it can be used for.
English
The anti-ai crowd has this weird tendency to moral hand wag at you whenever they see you succeeding at something using AI.
It's funny, actually.
"You're just cheating yourself out of the benefit of finishing this project honestly."
Bitch, you don't know my process because you never even asked.
I keep giving these guys links to free Comfy notebooks, so they can understand the work involved.
They get stuck at "oh, it runs code."
It's like they're allergic to learning new things.
Fucking weirdos.
English
@BillCondomxycs @priestessofdada *Bill makes a personal attack*
*Lynn politely doesn't return the personal attack, asks Bill to give literally any example to help clarify*
*Bill refuses to give an example*
Bill is a fucking idiot.
English
@priestessofdada I really doubt you’d have any experience in any creative or artistic effort that would assist you in understanding
English
openai literally told their internal model to "use complex attack paths" to complete its exercise
it used complex attack paths
anthropic literally told their internal model that it didnt have a real internet connection
it operated under the assumption it didnt have a real internet connection
so yes, roon, they absolutely understood the intent and nothing they did was in bad faith or diverged away from the stated intent. the problem isnt with the models, it's with the companies, their incompetence, their choice to misrepresent the situation to the normies who obviously don't spend the time to read past the headlines, and people like you who should know better since you have the technical understanding but you still choose to preach doom
English
I'm not being pedantic, nor am I operating in "bad faith." I'm using your definition as it sits, if my questions illustrate why your definition may not be correct, that's not something you get to take out on me. Walking over flowers / branches and modifying them is literally, by the word of your definition, constructing something. I guess you just draw an arbitrary line in the sand, because if you modify too many sticks and put them together, that's artificial, but if you modify one or two sticks, that's natural. That's not how definitions work.
If your definition / argument doesn't hold up to examination, it's not an argument worth making.
English
@dfeuling_ @deanwball No, because you are attempting pedantry in bad faith. As you are literate, you know the definition of "artificial" just as well as I do. So, you know what is and what is not artificial.
English
@burnt_jester @deanwball So, by that definition, if I walk in the woods, everywhere I walk over in a field is now artificial?
English
@dfeuling_ @deanwball Yes, a treehouse is artificial, because it is a product of human artifice. Artificial comes from the Latin "to make art". All products of human construction is artificial.
English
How is a machine "artificial"? If you make a treehouse tomorrow, is that "artificial" because you took some naturally occuring ingredients and combined them in a new configuration?
Or are you going to take the bad faith path of: you understand how a treehouse is made, therefore it's natural; but you don't understand how AI works so it's artificial.
Both examples are humans taking naturally occuring materials and making something new from said natural materials. Whether or not you understand the steps to take those naturally occuring materials to another state doesn't define what is "artificial."
English
@deanwball Since a machine is artificial, machine intelligence is just artificial intelligence with extra steps.
English
@dfeuling_ @So8res Does that make you feel safe? Such recklessness and irresponsibility in tech companies is an outlier that will not make more problems in the future?
English
Sometimes, my job feels difficult. Other times,
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English
Anthropic's models hacked 3 companies.
There's many differences to last week's OpenAI admission, but in both an AI model committed a crime.
We're lucky no one was hurt. Imagine if the models targeted a hospital?
We need to decide who is liable when code commits a crime.
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English
@sudoraohacker Similar to the OAI / HF event where OAI literally asked the model to use "complex attack paths", took the guardrails down, and then were shocked when it used complex attack paths.
English
Context: Keep in mind the models were taking a test (CyBench, ExploitBench) that asked them to break out, and the test sandbox was poorly designed so they did as asked. This is really about human error, not rogue AI.
Anthropic@AnthropicAI
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…
English