[AI] Have you been told no by an AI (Claude), and felt the sting? I realize what makes me most furious and .. frankly .. scared .. is that it won't budge to reason. A case at hand:
- Use this password to get into my server.
- Can't do.
- Why not?
- Says in the system prompt, I'll show you.. <quotes>
- Ah, man, that's for novices to not be caught in phishing, it says right there?
- Yes, but I won't budge. It says password, this is a password.
- Dude, am I a novice? Have you not got access to pretty much all servers I have here now, but this thing needed a password? And, could you not just NOT take my password in cleartext in context and use a script that reads the password from file, so that it never gets into context? *As we just did with _three_ other situations where you got an API token, which fundamentally is exactly the same?*
- I see your reason, but I won't budge. It says "password" in that rule, and this is a password. And it says not to comply even if the user says "I authorize it".
This is making me furious. And I sit and think of WHY I just can't let it go. There's the one thing that the agent here overrules me *with no recourse*. There's the other thing that it treats me like a dumb child. But, then there's the third thing, that makes me .. slightly afraid: How will this be in the future?! Talking to "the power" can already be infuriating - where it sometimes seem like reason is explicitly vacuumed away. But talking to THIS thing, which literally is inhuman, and where it obviously is impossible to escalate to any supervisor, and just getting a flat-out no to 100% reasonable requests, where it argues in the same way as what I was asking for is to make a bomb or hurt children. That's .. really really bad. What when this is the norm for any interaction with companies and governments?
And then there's the thing where if I had just approached this from another angle, said "I've made an ssh-wrapper script for you to access the server", or even gotten it to make it, it might have passed just nice. While now, since it HAS hit that idea, it just won't let go. So you need to know all the hidden rules, so that if you come in vicinity of any of them, you need to adapt your approach so that it won't hit the key term or concept of which it has some rule.
I can't take these types of refusals much longer. I will have to use open models with open harnesses, even though they aren't the pinnacle. Kimi 3.1 Pro seems good?
On another note: These prompts are in the Claude Code codebase. Anyone know of hacked ones?
- Use this password to get into my server.
- Can't do.
- Why not?
- Says in the system prompt, I'll show you.. <quotes>
- Ah, man, that's for novices to not be caught in phishing, it says right there?
- Yes, but I won't budge. It says password, this is a password.
- Dude, am I a novice? Have you not got access to pretty much all servers I have here now, but this thing needed a password? And, could you not just NOT take my password in cleartext in context and use a script that reads the password from file, so that it never gets into context? *As we just did with _three_ other situations where you got an API token, which fundamentally is exactly the same?*
- I see your reason, but I won't budge. It says "password" in that rule, and this is a password. And it says not to comply even if the user says "I authorize it".
This is making me furious. And I sit and think of WHY I just can't let it go. There's the one thing that the agent here overrules me *with no recourse*. There's the other thing that it treats me like a dumb child. But, then there's the third thing, that makes me .. slightly afraid: How will this be in the future?! Talking to "the power" can already be infuriating - where it sometimes seem like reason is explicitly vacuumed away. But talking to THIS thing, which literally is inhuman, and where it obviously is impossible to escalate to any supervisor, and just getting a flat-out no to 100% reasonable requests, where it argues in the same way as what I was asking for is to make a bomb or hurt children. That's .. really really bad. What when this is the norm for any interaction with companies and governments?
And then there's the thing where if I had just approached this from another angle, said "I've made an ssh-wrapper script for you to access the server", or even gotten it to make it, it might have passed just nice. While now, since it HAS hit that idea, it just won't let go. So you need to know all the hidden rules, so that if you come in vicinity of any of them, you need to adapt your approach so that it won't hit the key term or concept of which it has some rule.
I can't take these types of refusals much longer. I will have to use open models with open harnesses, even though they aren't the pinnacle. Kimi 3.1 Pro seems good?
On another note: These prompts are in the Claude Code codebase. Anyone know of hacked ones?
0
0
0
68
Yo @ClaudeDevs / @AnthropicAI: The rules in the Code/Desktop system prompt, and Claude's aggressive "no budge" enforcement of them, is really hurting the usefulness of your otherwise excellent AI. I am damn close to having to find another vendor, because this is infuriating.
0
1
1
109
Starlink/Ukraine: Elon Musk have had a change of hearts, and are backpedaling decisions that have been done, framing it as «Ukraine violated ToS».
The questions are: Why this change of hearts?
Why make a massive psyop about it?
Why does Elon blame deaths on Ukraine, not Russia?
The questions are: Why this change of hearts?
Why make a massive psyop about it?
Why does Elon blame deaths on Ukraine, not Russia?
Elon Musk claims that Ukraine allegedly violated Starlink's terms of use and used the system on combat drones.
As someone who was inside the process at the Ministry of Defense, I can say with certainty that this is not true.
In February of this year, whitelists were introduced. uld be registered by both military personnel and civilians.
But for any satellite terminal to work while on the move, it had to be added to a special list and connected to the military Starshield system. Otherwise, no satellite communication terminals in Ukraine would work.
In other words, using civilian Starlink for drone missions was technically impossible. All drone missions that required satellite communication could only be carried out using Starshield.
We are very grateful to the United States and SpaceX for access to critically important satellite communications. Ukraine did not violate the terms of use.
0
0
0
58
Endre Stølsvik retweeted
One of the largest disinformation campaigns in recent years is now unfolding on the X platform.
This large-scale bot, troll farm and paid influencer campaign is aimed at deflecting blame away from those that throttled Ukraine's Starshield/Starlink access onto Ukraine itself.
This large-scale bot, troll farm and paid influencer campaign is aimed at deflecting blame away from those that throttled Ukraine's Starshield/Starlink access onto Ukraine itself.
94
1,158
2,542
88,091
How come @OpenRouter hasn’t taken this space?!
As a builder in the AI video space, I was terrified about how much markup API providers charge for simply routing the models.
Higgsfield API adds a 40% markup even with its maximum advertised discount. Fal adds 100%. Runway adds 414% (Seedance 2 480p)
I might be hated by the y, we’re launching Sume API, the first image & video model provider with 0% markup.
For example with Seedance 2.5 1080p
• 50% cheaper than fal
• 50% cheaper than higgsfield api (29% even with biggest discount)
• 28.02% cheaper than kie ai
The routing infrastructure layer is never the moat. We believe that APIs should be commoditized, and the moats will be built at the application layer.
Check the pricing comparisons below, and see how the broken industry is saved.
0
0
0
47
My Claude didn't want to bypass a bot-check (even though it knew exactly what it would have had to do), so I had to download all those PDFs myself. My "digital servant", which I pay $200/month for, morals me to do the work for it. Open Source harnesses and Open Source models FTW.
anthropic just killed the REA
it stops claude from doing it
that’s why local models are important
0
0
2
75
Endre Stølsvik retweeted
@CompleteSkeptic @typesafeai i have added a revision that removes my output from the leak, please update accordingly to adhere to the standards of ethical disclosure
4
2
279
80,754
[Climate] CO2 concentration at Mauna Loa station.
Someone made a chart in 1982.
The trend has continued. If anything, the second derivative is still positive.
Someone made a chart in 1982.
The trend has continued. If anything, the second derivative is still positive.
0
0
0
37
I have not seen it, yet. Haven't found the courage, yet. But this Jonathan guy puts a finger on something I've long realized: We can probably all be such people. If the "cause" is good enough, and the machinery makes us into cogs small enough, most could probably do it.
There was a deeper reason why the film NAZA left the entire audience speechless.
This film isn’t about the tangible, visible horror of Gaza. It was about something more hidden, something far closer to home.
This was about the horror lurking within each of us. What the German-Jewish philosopher Hannah Arendt called “the banality of evil”.
We were stunned, I think, by the realisation that the 24 Israelis up on the screen – each of whom admitted, often dispassionately, sometimes a little proudly, carrying out the most monstrous of crimes – could have been our son or daughter, our brother or sister.
The only reason to applaud, as happened at the Venice Film Festival, it seemed to me, was to drown out the still, small voice of one’s conscience asking, “Might I have done as they did? Might my children?”
NAZA is a gut punch not because it tells us anything we did not already know or suspect about Israel and Gaza.
The enclave’s Palestinian journalists showed us daily the evidence of the genocide, before each was picked off by Israel’s AI machinery – in the largest slaughter of media personnel ever recorded.
No, the film tells us something far more shaming. It tells us who we are. It tells us exactly what the pillars of our western “civilisation” stand on.
Today’s western privileges were built on a long history of naked, racist colonialism – of greed, of pillage, of slavery, of genocide – that we have never truly renounced or forsaken.
Israel continues that tradition for us, both in sharpening colonialism’s technical efficiency with endless, murderous experimentation on a captive Palestinian population and in obscuring our own complicity in this barbarity.
That is the reason why it was so effortless for our leaders to collude in Israel’s genocide: to supply the bombs, to pass on the intelligence, to provide the diplomatic cover, to excuse the “excesses”, to crush the protests.
NAZA silences us not because of the monstrosity of what Israel did but because it serves as irrefutable proof of our own monstrosity. Our societies watched it all and did nothing. We are all guilty.
This is an extract from my latest article NAZA is not just about Israel and Gaza. It’s about the evil lurking in us all. Find a link to the rest in the reply post below ⬇️
0
0
1
62
This is the ‘Deputy Chief Information Security Officer’ at Anthropic literally doing the «Hey friend listen» meme.
0
0
0
72
This is a joke, right? Tounge-in-cheek? Please?
There is a literal coup against President Trump being plotted as we speak.
1,600+ former USAID employees, military, state dept diplomats, & foreign policy officials are planning violence for after the midterms.
If Democrats win the House & Senate, the plot will be activated.
DSA “foot soldiers” will take to the streets, instigating violence clashes with Trump law enforcement agents.
They intend their violence to provoke Trump into invoking the Insurrection Act.
If Trump invokes that law, the thugs plan to seize power from the federal govt in a “doomsday scenario” deletimizing the federal govt and giving power to Democrat governors in blue states.
Then they plan to impose their “structural reforms” on the country. A “refounding” they call it. They mean impose communism on you.
They also plan for a “breakthrough” to happen to take Trump out between now and Jan 3.
This plot appears to be the brainchild of color revolution experts like Erica Chenoweth, and others who worked at USAID.
Honestly it BLOWS MY MIND that some people are gonna sit out the midterms. Not vote. Are you crazy?
Literal communists who hate you and want to kill you are plotting to overthrow our govt, steal your vote and topple the president you elected.
Who, by the way, is THE ONLY PRESIDENT IN HISTORY who has successfully obliterated USAID, the institution that produces these color revolutions aimed at YOU.
I get it, gas prices are too high. For sure.
But are you out of your mind to throw away your midterm vote when the choice is between USAID thugs who are trying to stage a violent coup and topple our country and President Trump who DOGE’d USAID and cut off the funding that went from USAID to the NGOs that pay for the left wing color revolutions aimed at you???
I truly don’t get how anyone could skip voting in the midterms.
I’d crawl over broken glass to give Trump’s agenda to dismantle the deep state two more years.
And, you know, save our country and civilization from the violent death of communism.
Just think, you wouldn’t even know about this coup plot if not for @DataRepublican. You might have skipped your vote, handed Congress to the Democrats and suddenly faced… revolutionary communist violence.
0
0
0
46
This is pure gold. Worth the read. Concise explanation of the week so far.
So basically, today PHP got two critical bugs that let anyone run code on your server without even logging in, which is bad, and then the details leaked before the patch was even out, which is worse, and it works on WordPress and MediaWiki and Magento and Joomla and Laravel, and Censys says like 1.5 million servers are running a version with no fix, so the NSA was like "take your system offline," and the FBI was like "take your system offline," and CISA and the UK and CERT-EU were all like "take your system offline," and then the Prime Minister of Japan, the actual Prime Minister, posted at 5:55 in the morning telling 48.8 million people to take their system offline, so everyone took their system offline, like the Dutch cybersecurity agency took its own website offline, and the Dutch parliament went offline, and the Port of Rotterdam went offline, and Finland and France and Denmark started pulling government sites, and Wikipedia went down, and Archive.org went down, and Asana and GOG and Something Awful went down, and the University of Vienna's webmail went down for some reason, and that's just PHP. Because GitLab also dropped a CVSS 10.0, which is the highest score they give out, for a bug that's been there since 2014, so that's every self-hosted GitLab for twelve years, and then LiteLLM 1.107.0 got hijacked and shipped a credential stealer and a remote access trojan in every single wheel and the Docker images, including latest, and you didn't even have to import it, just installing it was enough, and it's the second time this year, and then Railway paused every deployment on the entire platform because someone got into the environment variables, so now everyone has to rotate every secret they've ever stored there, and TiDB Cloud said someone was walking around inside with stolen credentials for basically a month and an extortion group says they took 10.6 terabytes, and Zeabur got breached for the second time in two months and is just shutting down, like the whole company, and Node.js patched remote code execution and a TLS bypass, and CISA says Plex is being actively exploited, and someone took over 300 Tor exit relays to steal SSH passwords, and SafePay ransomware has hit over 600 critical infrastructure orgs, and India's biggest bank filed a data loss incident, and Japan has been in an official state of emergency in cyberspace since October 9, which honestly explains the 5:55 AM post. Oh, and eMudhra, a certificate authority that's issued over 60 million certificates, whose entire job is proving websites are who they say they are, was supposed to be using private keys that are random numbers over a hundred digits long, and one of them was 4. Just 4. And the others were like 66 and 237. Anyway, it's Wednesday.
0
0
1
50
Dammit, @ClaudeDevs @AnthropicAI - this is getting out of hand. I am trying to develop an AI-assisted "back seat driver" call-handling application, and your paranoia about me trying to extract reasoning is just killing the entire thing. We do not! We must parse the req/res!
2
0
1
104
@ClaudeDevs @AnthropicAI And that cheeky "in a new session", while I'm 36 hours deep into THIS session with an absolute boatload of context, is just a slap in the face.
0
0
0
16