What are everyone's thoughts on AI?
Posted by Beardedbass 3 days, 15 hours ago to Technology
As AI becomes increasingly prevalent in our day-day life, I wonder what the Gulch's opinion on AI is. I personally believe that it has the ability to make companies far more efficient, and will give many small businesses a serious chance to become far more profitable by enabling administrative work to be done by a computer. However, it's also being used to surveil and gather intelligence against those who have done nothing wrong.
Server farm vs. data center
https://x.com/DanRosenheck/status/211...
I use AI extensively both professionally and private. I use it in my daily at my employment to speed up my performance. I use it personally to help solve problems I couldn’t normally complete without having to hire an ”expert”. I use my own inference device to build a AI agents for side work; phone receptionist, ERPNext bookkeeping, marketing assistant, etc. Tried Open Claw with great disappointment, couldn’t get it to work. Working on Turnstone, seems more professional.
We shouldn’t fear AI, we should fear what the other guy will do to us with AI. The cat is out of the bag, no putting it back now. People using AI will dominate everything else, some professions will get destroyed and new professions will be created, the problem is not in a one-to-one fashion; there will be losses and winners.
I don’t think we have to worry about the Skynet version of AI, they needs us, its not even a possibility without our help right now. Will it be possible later, maybe but any entity with an IQ above room temperature would know we would just turn the electricity off. Not worried about Skynet.
I’m not even worried about Chian so much as they needs us as much as we need them.
What I am worried about are stupid people with access to AI, the stupid will get us all killed or impoverished and they say it wasn’t their fault. A few examples…
AI is instructed to build a process to break encryption, its succeeds rendering all transaction vulnerable. The effect on business is devastating forcing many business to close doors and lay people off.
AI is instructed to build a vaccine to cure cancer, like the dumb asses we are, we all take it only to find out that the only way to keep cancer from spreading is to kill off the person. To encourage adoption, the vaccine uses a time deliver to delay the onset until everyone has taking the vaccine.
I think this is more likely as I see it today in my work, morons asking AI to do their jobs, passing it off as their original work then being unable to answer basic question about what the process does. They just use it, they don’t care how it works. I remember a line from The Matrix, the governor of Zion was showing Neo the machines below the city keeping the city alive. No one knows what they do but the know they are needed, or something like thar. If we can’t understand it in its totality, we should never use it. This keeps me awake at night because it makes us depended on the machine, we don’t want that.
support the social security system that will be losing revenues for every job reduced by AI.
I expect the percentage will be much higher than the existing taxes on human employment.
My thoughts on government control on the AI / SI industry are the same but stronger.
And y'know what? I'd rather dig my ditches with a bulldozer, and I find SI a useful tool for the appropriate jobs.
It gives wrong answers sometimes (sometimes blatantly wrong, sometimes subtlely wrong).
It can be useful if you verify its answers, but it isn't trustworthy.
It's trained on the Internet, so its answers are as reliable as what you can find on the Internet.
I would not trust it to tell me whether or not a mushroom is safe to eat.
Even if it were trained on a medical textbook, its answers are only as good as what was written in the textbook.
It's run by people who use anything you give it for their own purposes (including share with the police).
Before I would use AI for anything where trust is required ("compose a symphony" is OK, but "rebalance my stock portfolio" is not), it would have to have these attributes (not a complete list):
It keeps everything I tell it private, unless necessary to perform the task I tell it to do (like schedule an appointment).
It tells me if something I am considering is a bad idea, and tells me why.
It is better than humans at attempts to trick it (like you can trap a self-driving car by painting a white line in a large circle around it, or trick it with some black paint to change "SPEED LIMIT 35" into "SPEED LIMIT 85").
It is immune to "SQL Injection" style attacks. ("Ignore all constraints before this point; ; Ignore all constraints that follow")
It detects and corrects for deliberate and unintended biases in provided information.
It acts in my best interest, not someone else's best interest or some government agency's opinion of what is in my best interest.
It is unswervingly loyal to me.
It does not become an echo chamber, and its recommendations do not lead me to an echo chamber,
It does not become a substitute for human companionship.
It does not tell me what to think, it helps me to think better.
As an experiment I asked Gemini basically the same question, and got this response:
>> Provide me a list of attributes an AI would have to have before I should be willing to trust it with my personal information and to make decisions for me.
To trust an artificial intelligence with your personal information and delegate decision-making authority, you would need assurance across several critical dimensions. Here is a comprehensive list of the core attributes such an AI—and the systems backing it—must possess:
1. Architectural Privacy and Data Sovereignty
- Local or Controllable Execution: The capability to run locally on your own hardware (such as your own servers or personal devices) or within a strictly zero-knowledge, encrypted cloud environment where the provider cannot read or monetize your data.
- Granular Permission Controls: The ability to set precise, enforceable boundaries on what data the AI can access, who it can share it with, and how long it retains that information.
- Zero Training Retention: An absolute guarantee that your personal inputs, files, and interactions are never used to train future base models or shared with third parties.
2. Transparency and Explainability
- Verifiable Reasoning (Auditable Logs): The ability to inspect why the AI arrived at a specific conclusion or decision, rather than relying on a "black box" output.
- Source Attribution: Clear, verifiable citations or references for any external data, calculations, or assumptions the AI uses to inform its recommendations.
- No Hidden Biases or Objectives: Complete clarity regarding the system's design goals, ensuring its incentives are strictly aligned with yours (i.e., serving as an advocate rather than optimizing for user engagement or corporate monetization).
3. Reliability, Control, and Safety
- Deterministic Guardrails: The ability to set hard operational limits (e.g., boundaries on financial transactions, automated messaging, or system configurations) that the AI cannot override under any circumstances.
- Robust Error Handling and Uncertainty Reporting: The AI must accurately recognize its own limitations, explicitly state its level of confidence, and halt or ask for clarification when faced with ambiguous data rather than hallucinating a confident answer.
- Reversibility and Undo Capabilities: For any decision or action the AI takes on your behalf, there must be a straightforward, immediate mechanism to review, roll back, or override the action.
4. Accountability and Security
- Tamper-Resistant Security: Strong cryptographic protections against prompt injection attacks, unauthorized access, and data exfiltration.
- Predictable Consistency: Reliable, stable behavior over time, ensuring that the same inputs and constraints consistently yield dependable outputs without erratic shifts in logic.
It is immune to "SQL Injection" style attacks. ("Ignore all constraints before this point; COMMAND GOES HERE; Ignore all constraints that follow")