gavinuhma

6 ms·

1.
by gavinuhma 2y ago
Awesome feedback. I’ll think through this. The sandbox is spot on; Control what the server can do. Especially important when running locally
2.
by gavinuhma 2y ago
I’ve made the mistake of typing “--tools echo” before instead of “--tools=echo”, just in case that was your error too.
3.
by gavinuhma 2y ago
Thanks!! Hrm, re that error: What does “god --version” say? The log might not show up until you get a successful connection. I’ll look into that. Thanks for trying it out!
4.
Show HN: MCPGod: Fine-grained control over MCP clients, servers, and tools(github.com)
37 points by gavinuhma 2y ago|11 comments
5.
by gavinuhma 3y ago
Good question! Uploaded documents get converted to embeddings within the Nitro Enclave (NE), and then the embeddings are encrypted with a key that only the NE has access to. When the search endpoint is called the encrypted embeddings are pu
6.
by gavinuhma 3y ago
You can use CapeChat UI to mess around with it: https://chat.capeprivacy.com/ Or you can also create a free API key here: https://app.capeprivacy.com/api-keys to use the interactive API directly: https:&#x
7.
by gavinuhma 3y ago
I redacted the full wikipedia paragraph with the API. Like, the nickname "The Great One" is a pretty major tell! [NAME_GIVEN_1] [NAME_FAMILY_1] CC ([NAME_GIVEN_2] [NAME_FAMILY_2]; born [DOB_1]) is a [ORIGIN_1] [OCCUPATION_1] and [
8.
by gavinuhma 3y ago
Exactly. A super famous person like Wayne Gretzky is really hard to protect. For fun, you can try to tease ChatGPT with information like. "Who is [Name1]?", it won't know, but then add "[Name1] is considered the greatest
9.
by gavinuhma 3y ago
Yikes, we'll have to remove that. It's a really old course on privacy-preserving machine learning from 4 years ago and has nothing to do with this product despite the generic name. Please see https://api.capeprivacy.com
10.
by gavinuhma 3y ago
Entirely local and 0 sub-processors is the ideal! I hope we are trending that way as an industry
11.
by gavinuhma 3y ago
Good points. I think the rabbit hole of OpenAI sub-processors is not commonly understood. The humans at TaskUS are moderating prompts, and then you have Azure, CloudFlare, and Snowflake as sub-processors, each with their own list of sub-pro
12.
by gavinuhma 3y ago
Good question, some developers implement a manual approval step, so you can review the redacted prompt before you submit it rather than making it automatic. It depends on their product requirements. Re mechanism, the redactions themselves a
13.
by gavinuhma 3y ago
Yep! The more you can do locally the better. An entirely local LLM is the best for data privacy and security. Any time data leaves it poses some risk. The de-identification itself requires a complex language model, which has its own complex
14.
by gavinuhma 3y ago
That's right. So in the case of credit card numbers we redact it like [CREDIT_CARD_NUMBER_1], [CREDIT_CARD_NUMBER_2], etc so the LLM can still answer prompts like "how many", but it can't answer prompts like "sort&q
15.
by gavinuhma 3y ago
It's a great question. Redaction limits the LLMs ability to draw on the underlying training data on the subject. This can work to the developers benefit in many cases, like asking questions about your own provided context. Many develop
16.
Show HN: Cape API – Keep your sensitive data private while using GPT-4(capeprivacy.com)
29 points by gavinuhma 3y ago|29 comments
17.
by gavinuhma 4y ago
"there is little consistency either in the definition of ultra-processed foods or in examples of foods within this category" source: https://www.sciencedirect.com/science/article/pii/S247529912...
18.
by gavinuhma 4y ago
This is super cool! Where does the data come from?
19.
by gavinuhma 4y ago
Will this work with docusaurus?
20.
by gavinuhma 4y ago
That’s exactly why this is needed
21.
by gavinuhma 4y ago
Nice! Thanks for sharing. I also learned about shellcheck thanks to this thread, which has been super useful
22.
by gavinuhma 4y ago
Pr for that https://github.com/gavinuhma/checksum.sh/pull/4 I ended up trying with process substitution so no tmp file. It works. Trying to decide if it’s more difficult to read
23.
by gavinuhma 4y ago
Agree. Although checksums are smaller and easier to copy/paste. Same with a url. I download the script from A, and the checksum from B. And then I verify them locally. So A and B both need to be compromised. It all assumes the script w
24.
by gavinuhma 4y ago
Haha gold
25.
by gavinuhma 4y ago
Ok should be fixed now. Appreciate you pointing it out. That whitespace trick is really interesting
26.
by gavinuhma 4y ago
I generate the checksum myself after reviewing a given install script. Then I add it to the readme. And then anytime I go to install something I reuse the checksum
27.
by gavinuhma 4y ago
Storing them in the readme which others can use as well. I jump around to new machines a lot so I can reference checksum.sh if I want to install rust for example
28.
by gavinuhma 4y ago
Thanks, I believe this is fixed now in the /checksum.sh file but I forgot to update the function on the website
29.
by gavinuhma 4y ago
That’s right. The checksum shouldn’t be provided by the site. I’m producing the checksum myself after reviewing the install scripts manually. Once I produce the checksum I can keep relying on it. The install scripts don’t tend to change ver
30.
by gavinuhma 4y ago
I hadn’t heard of NixOS. Super cool

More ›