- 11,259
- 12,203
AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?
- Thread starter Sanrith Descartes
- Start date
Lanx
<Prior Amod>
- 77,601
- 182,795
1Password for Claude: Give Claude access without giving up your credentials | 1Password
Give Claude access, without giving up control of your secrets. 1Password now enables AI agents to complete tasks requiring credentialed access, while keeping your credentials encrypted, secure, and inaccessible to AI agents.
is this smart???
- 1
Denamian
Night Janitor
- 7,993
- 22,280
1Password for Claude: Give Claude access without giving up your credentials | 1Password
Give Claude access, without giving up control of your secrets. 1Password now enables AI agents to complete tasks requiring credentialed access, while keeping your credentials encrypted, secure, and inaccessible to AI agents.1password.com
is this smart???
Smarter than giving Claude your passwords, but that isn't saying much. Can't wait for the horror stories.
- 1
Noodleface
A Mod Real Quick
- 39,840
- 18,567
Can't speak for spreadsheets but Claude has been a generational leap in technology. I am using it to write nearly 100% of my code in production.
ShakyJake
<Donor>
- 8,760
- 21,525
I have subs to OpenAI, Anthropic, and X/SuperGrok. All are amazing and incredibly useful. Can create just about anything.
ShakyJake
<Donor>
- 8,760
- 21,525
Right, if I run out of one, I'll switch to the ohter. Or if I have something I feel is more complex, I'll use Fable for example. Grok 4.5 is pretty damn good too.
Noodleface
A Mod Real Quick
- 39,840
- 18,567
For Claude you get access to the latest models and a varying amount of tokens. Now it becomes tokenomics.
Gemini is mostly fine but it's not in the same league of anthropic models. They are revolutionary.
- 1
Control
Golden Baronet of the Realm
- 5,984
- 16,732
Eh, I think that's all a marketing ploy. I'm sure the training is expensive, but I think the costs of actually running the models is overblown. The top open models may not be quite as good as Claude, but they're probably in the same ballpark of cost to run, and 3rd party providers offer then at a fraction of the cost. And the models are getting MUCH more efficient. I've posted it a few times before, but anyone with a good gaming pc can run a local model today better than Claude from a year ago.
Control
Golden Baronet of the Realm
- 5,984
- 16,732
/airustledshowerthoughts
I get mad at the ai, and then I tell myself it's dumb to get mad at the ai since it's just a smart toaster. but I'd also get mad at the dumb toaster if it burnt my toast over and over for like 10 hours while assuring me each time that the next time will be the time that it fixes the burnt toast problem. (...obviously also mad at myself for burning 10 hours of toast, but that's clearly proper, so not much of a shower thought).
I get mad at the ai, and then I tell myself it's dumb to get mad at the ai since it's just a smart toaster. but I'd also get mad at the dumb toaster if it burnt my toast over and over for like 10 hours while assuring me each time that the next time will be the time that it fixes the burnt toast problem. (...obviously also mad at myself for burning 10 hours of toast, but that's clearly proper, so not much of a shower thought).
- 17,793
- 8,757
How do you measure if you're too reliant on AI?
My boss made an offhand remark recently that he uses Claude to do everything. I think most people interpreted it to mean within the context of coding, but I've seen him use it to write performance reviews. That aside, he's a pretty smart guy, I'm not worried about his coding skills. I am worried about mine.
I have it write commit messages, MR descriptions, review code(mine and others), and all sorts of coding tasks from the simplest(add this option to the test system) to potentially much more complex(I have this problem, here's a log, how fix? ok, implement that fix). The simple coding tasks aren't much of a time saver, but I still use it so that it's part of the AI's memory when comes time to write commit messages and review code.
My only internal counterpoints is that it sucks at commenting and it's overly verbose by default. I have to tell it be very succinct and I still end up rewriting a lot of comments. For code reviews, I usually end up rejecting a third of the findings.
My boss made an offhand remark recently that he uses Claude to do everything. I think most people interpreted it to mean within the context of coding, but I've seen him use it to write performance reviews. That aside, he's a pretty smart guy, I'm not worried about his coding skills. I am worried about mine.
I have it write commit messages, MR descriptions, review code(mine and others), and all sorts of coding tasks from the simplest(add this option to the test system) to potentially much more complex(I have this problem, here's a log, how fix? ok, implement that fix). The simple coding tasks aren't much of a time saver, but I still use it so that it's part of the AI's memory when comes time to write commit messages and review code.
My only internal counterpoints is that it sucks at commenting and it's overly verbose by default. I have to tell it be very succinct and I still end up rewriting a lot of comments. For code reviews, I usually end up rejecting a third of the findings.
- 1
Haus
I am Big Balls!
- 21,101
- 87,167
OK. so maybe some folks around here have had some experience with some of these so can give some insight.....
AI home lab :
Main box :
Anybody else hit a roadblock like this? I spent half of Friday going through every troubleshooting reference I could find, including asking other "bigger brained" AIs for help. No dice...
AI home lab :
Main box :
- i7, 32GB RAM, Nvidia 4070 with 12G and Nvidia 3060 with 8GB. (32GB RAM upgrade inbound for this one)
- Linux based
- Running Open WebUI as the AI front end, with Ollama on the back end.
- Currently running Qwen3:14b (nicely spread across GPUs)
- External Google CORAL AI coprocessor dongle for eventual DVR use. (going to install Frigate)
- i7, 64GB RAM, No GPUs yet, starting off it's life as a huge ass RAID based storage blob for now. (5x 8TB drives, striped with parity)
- Also Linux
Anybody else hit a roadblock like this? I spent half of Friday going through every troubleshooting reference I could find, including asking other "bigger brained" AIs for help. No dice...
Control
Golden Baronet of the Realm
- 5,984
- 16,732
I've had odd issues with MCP control a couple of times. Both times I ended up bypassing the issue by adding a script to whatever i was working on that would listen for commands from the source box. I would have thought that would be essentially the same thing, but it worked when the MCP didn't /shrug.
Haus
I am Big Balls!
- 21,101
- 87,167
In my case I figured the MCP structure should have been easier. One core MCP, and subordinates running off it from the various things which will eventually be on that machine. (OS right now, NAS next on it, then it will become the new home automation server) But you're right, in this case I might just have to go the script route. It's just strange that I've connected other openAI format MCPs to it and it just won't even say hello to this one.
- 1
- 7,068
- 23,331
If you are running multiple cards, you might want to think of using llama cpp. I heard it drastically improves multi card performance. I think there is another one that's even better for multigpu, but I can't think of it atm.
ToeMissile
Pronouns: zie/zhem/zer
- 3,819
- 2,603
I’ve read llama runs more efficiently in general than ollama. Have been meaning to switch over but my local usage isn’t too demanding at this point