Python
Securing LLMs and Chat Bots
How to secure against prompt injections and jailbreaking LLMs and chatbots with prompt protect.
What were the other versions of the Web?
Web 1.0 - Content by companies
Web 2.0 - Content by users, Social Connections, everybody hates it, it destroyed society, and ten people own it
Web 3.0 - Scammers telling you money's dead, while asking you for money
Web 4.0 - Burning it all down and starting again
Nearly every colleague I know in the data science industry has commented on ChatGPT
Many talk about how it's not that hard to do, (it is hard, especially to get to the quality of a conversation we're seeing).
The folks at OpenAI did several amazing things
-
They Launched - that's huge
-
The general public was blown away - Huge'er
-
They made Google flinch - HUGEST
Googles BRAD faltered after a hasty launch in response, proving how amazing OpenAI's launch was.
Launching is hard!
But ChatGPT is not a search engine.
Search engines depend upon people creating content,
-
Search engines indexing that content
-
Take what users are looking for and compare it against that content
-
Present results to users in an order that will provide them with the most authoritative results
-
Or an order that will most likely generate a click through for that user. Those orders are often very different.
The business model however is simple, show Ads, give the user what they want, they will come back, show more Ad's and make more money.
Then "capture the behavior" of users coming to the search engine, and display sponsored Ads ahead of results, make lots more money!!
Easy enough right?
Google's combined Ad revenue is about ~$220B p/a (2022), so they've figured it out.
But what happens if you don't need a search engine?
What if you can provide a user with content directly effectively killing off the need for a search or even a click through?
For content creators, it means less clicks, less eyeballs, less Ads and less revenue, a battle that today they are currently waging with the hand that feeds them.
Search engine companies have for years attempted to develop Zeitgeist projects, something to predict the needs of a user and provide it to them, be it snippets, default results, current affairs etc...
A lot of it though was Yesterdays News Tomorrow
Stable Diffusion impression of a Zeitgeist Spirit
Google was a bit better populating the top section of a search results page with realtime data, or sponsored shopping results, business listings, Q&As, reducing off-site clicks by as much as 30%.
Often hurting the content creators or businesses who would have normally received those clicks.
See Yelps commentary on Googles anti-trust fine in Europe https://www.nytimes.com/2017/07/01/technology/yelp-google-european-union-antitrust.html
With LLMs / Generative AIs; we're now at the point where content can be generated on demand, there's a huge leap to make it valid, trust worthy, and timely but those are problems we've solved at scale over the years. (Unless you're Facebook or Twitter)
Generative AI is moving the needle from Predicting a Users Needs, to Tailoring Precisely to their Asks.
What happens now to content creators when they're competing with something that can generate and adapt content specifically for that end user on any topic they're interested in?
With an adaptive tone, bias, or saliency that a system has identified from past behavior, all tailored to maximize your experience and satisfaction while achieving a target.
And if they're not happy, can click "Try Again"
Beyond clickbait, favorite sites, creators or narratives, Generative AI will change the business model of the internet and possibly kill internet advertising as we know it.
But does it also kill search engines?
Stable Diffusion impression of businessmen fighting
We know the answer for right now is no, but for how long?
There are four things that Generative AI has to do, outside of the validity, trust worthiness and fixing hallucinations before it can deal a death blow to the current bastion
-
Expand to verticals, that is a chasing game, one that is a lot of fun beat X at Y.
-
Recreate the form factor, it's no longer about being the Gateway to the Internet
-
Google is that gateway, and dominated by also being the browser
- And the phone outside of the US at least
- And capturing all the profitable verticals
- And being a massive data broker / hoarder
- MSFT Bing tried with IE / Edge and by being the Desktop but haven't succeeded
-
MSFT will do a Cortana again and make it stupid, they can't help it
-
-
-
Generative AI needs to be in every house, pocket, car, business and school to succeed
- For that to happen, Generative AI needs to deliver upon the expectation of users for the last 50 years.
-
Determine a new business model for the internet that does not have a gateway and revenue generator combined.
- Content attribution, inspiration and rights will eventually have to be encoded into LLMs and Generative Art AI
- This will lead to asks for rev share and copyright determinations (currently Getty is trying)
- Content generators will also need to find new model outside of display ads, hopefully it's not new mediums that frustrate users.
- The platform model was successful until Google kept killing it https://www.theverge.com/2019/12/16/21020323/youtube-policy-harassment-cyberbullying-drama-pewdiepie-steven-crowder
With point #3 & 4 there is a quote as it relates to Google
“If an injury has to be done to a man it should be so severe that his vengeance need not be feared.”― Niccolo Machiavelli, The Prince