zer0c00ler

Wild that this about behavior from 7 years ago - imagine additional fines that will need to be paid in 7 years for current market manipulation by some of these giant players (incl in US). Things are moving really slow. The amounts are also just a drop in the bucket...

Splitting up Google seems more and more likely, probably would raise the overall market cap and drive innovation

A lot of people wrongly believe this means no firewall. If you don't use firewalls (yes, even for web traffic and SaaS, limit who can access your stuff), you are doing something wrong.

For anyone who didn't know Google apparently deleted (like in making it unrecoverable) a larger customers subscription (and GCP data) and also all backups as far as it sounds. The customer luckily had their own backups elsewhere outside of GCP.

That's insane.

Just out of curiosity, do you play an instrument?

I'm wondering if it impacts people with strong emotional bond to art more than others.

I felt it was just super weird and nonsensical.

Whenever this story comes up I think of this movie, called 23, that's tells the story from the hackers perspective.

https://en.m.wikipedia.org/wiki/23_(film)

Both stories are fascinating.

Microsoft's mice and keyboards also were exceptionally good

I think each test served exactly its purpose and it's incredible to see the rapid progress. Unclear why you believe the first two were not a success at all? Did you expect a novel vehicle like this would just work the first time?

Google mitigated it via CSP , as did Bing Chat.

ChatGPT is still vulnerable btw

Agreed mostly, although I'm less of a fan of the "dream" terminology.

The key takeaway is that we can't trust LLM output and systems need to be designed accordingly.

Eg Agents will never be able to securely take actions autonomously based on LLM responses.

This sounds good, but didn't Apple get off the hook?

Anyone can decipher what the difference are?

Multiple resultsets is a common term. Where a resultset is a table basically.

SQL Server (since 2005) can actually even have multiple active resultsets on a single connection. Assume same is true for most other database servers.

What are the reasons that makes this possible? The articles I have seen are not explaining what makes this different? What needs to be invented to get end to end security for push? I might be missing the obvious, apologies upfront.

Likely just 0 testing.

Today I switched Outlook to the new Outlook and then it couldn't access my email account because of some licensing issue? No other error or how to resolve.

Who allows things like this to be shipped without minimal QA is beyond my imagination...

Read OpenAI's response to this security issue carefully - it tells you a lot about how they think of being responsible for issues like this. I remember they put all the blame on the open source library, rather than taking responsibility themselves.

I have trouble understanding what Q is. It can't be that new Chatbot named Amazon Q in AWS, because it has very few capabilities. It's kind of embarrassing.

What am I missing? Is there something else shipping in future?

doubt any of the media outlets actually used it, they are just repeating Amazon's statements

I have trouble understanding what Q is. It can't be that new Chatbot named Amazon Q in AWS, because it has very few capabilities. It's kind of embarrassing.

What am I missing? Is there something else shipping in future?

One observation is that the endless thank you and love you messages are getting a bit tiring... even Twitter likes show a steep decline in interaction and interest.

Can only play that card so many times.

Yeah, OpenAI lost a bit of its magic. It's sad because it was really fun so far to see all the great progress.

But there are so many unanswered questions still and the lack of transparency is an issue, as is the cult like behavior that can be observed recently.

Wouldn't open source mean you can recreate the model from scratch by knowing the training data?

There is a weird definition of open source that creeps in here.

Llama2 is effectively, proprietary compiled software. GPT-4 is proprietary compiled and hosted as SaaS.

We have yet to see a true open source LLM...

AI/ML is reinventing so many terms it's a bit annoying.

Elon brought him in, which is quite the irony. Funny even. It also is the reason Elon and Larry Page don’t get along anymore.

Ilya is certainly world class in his field, and maybe good to listen to what he has to say

Three engineers isn’t a lot honestly after such a stunt. I’d assumed there would more loyal folks, but maybe most are really in for the mission.

The next couple of weeks will tell.

You assume they were indeed blindsided, which I very much doubt.

I think it’s a good outcome overall. More decentralization and focused research, and a new company that focuses on product.

It’s 195 countries!

I doubt this was a surprise to them, I’m sure Sam was well aware of the concerns and repeatedly ignored them, and even doubled down. Putting OpenAI’s mission in jeopardy.

Many politically aligned folks will leave, and OAI will go back and focus on mission.

New company will emerge and focus on profits.

Overall probably good for everyone.

His timeline.

I don’t think there are 230+ countries, at least not on this planet.

No, it’s 11:24 pm pst and not available for me.

ChatGPT and plugins were already agents. An agent just does stuff on somebody else’s behalf. Browsing, browses the Internet for the user, code interpreter creates and runs Python code for the user…

What you describe are what’s called autonomous agents, and i agree that I also expected some interesting announcement there but probably too many security issues.

Yes, thats also why policies aren’t enforced. I assume OpenAI knew all along that plugins probably are a dead end. Wonder if they will be around 6 months from now.

But Actions will have all the same security challenges basically.

Yes, a prompt injection will persist longer and can have impact later in a conversation for instance.