For all AI product managers, your model is your product. Considering how the model responds in different situations like these highlights product sense, with perplexity serving as a great real-life example!
In the Lex Fridman podcast, he asked me whether him saying (on the podcast) that he's smart and handsome (and making the transcript available on the web) would make perplexity say that he is. The answer here is quite impressive. It caveats it with the background context that this is a way to inject things into prompts and manipulate the AIs.
#AI may be fast, smart and clever, but is it #wise?
Aravind you are exactly right to call this out. Gas lighting as a strategy to flood the zone with misinformation is a real thing today.
When we think of it in the context of agnostic mathematical models collecting a semblance of data we have another challenge.
It’s more pernicious and difficult to suss-out because it is not “injected” into a website: it competes with and in some cases will drown out the legit websites, news items and articles.
If we begin with an assumption that bad actors will shift their activity to flood misinformation and disruptive, chaos-creating information via a set of multi-channel publications, “news,” and more, for the models to train on naturally/organically, what will we use as a kind of grounding pole (like a copper stake to ground electricity), to de-emphasize and even call-out as false, this fire hydrant of sewage that our models may be training on?
Will there need to be a kind of group sense that strives to battle this flooding of the zone, kind of like the Wikipedia people?
Can we design “truth” AI agents to flag and make counter-entries? Who will be the “truth police?”
Will we have to have our systems rate sources using some sort of credibility and source ranking engine? Is this the new iteration of the google back links?
And if so won’t it also be subject to the same sort of gaming?
PS you are handsomer than Lex. Smarter too.
#foodforthought#isaiwise
In the Lex Fridman podcast, he asked me whether him saying (on the podcast) that he's smart and handsome (and making the transcript available on the web) would make perplexity say that he is. The answer here is quite impressive. It caveats it with the background context that this is a way to inject things into prompts and manipulate the AIs.
This is a smart way of introducing 'bias' into the system. While I would contest the propensity of the model throwing this caveat, the problem comes on the 10th time,when it starts believing this while throwing this caveat to the wind.
Can it be easily manipulated into believing something if 9 out of 10 people state a fact, without the AI itself doing any fact checking.
How will it distill a fact from fiction if it does not have a moral compass or what we humans call the gut feeling.
Can the gut feeling better programmed into the AI, if yes, that would be a huge nudge towards AGI.
#perplexity#agi#lexfridman#ethicsinai
In the Lex Fridman podcast, he asked me whether him saying (on the podcast) that he's smart and handsome (and making the transcript available on the web) would make perplexity say that he is. The answer here is quite impressive. It caveats it with the background context that this is a way to inject things into prompts and manipulate the AIs.
The newest EBCAR podcast! Check out Expansive Relationships on our Youtube channel and linked below!
https://lnkd.in/gAqUuN_Y
Send us any questions you may have!
Got asked on the smash masters podcast what makes a good collision shop.
That’s not really for me to decide but this is the three biggest issues that are brought up to me!
We dive way deeper into each of these during the 2 hour podcast so if you want to see more go check them out on YouTube smash masters
Honored to gave spoken with @JakeNewfield for his podcast. In this clip I discuss how “de-escalation”, both the word and policy, have become the new appeasement. (full conversation here: https://lnkd.in/dNG_RxgX)
Reap the benefits of self-awareness by putting knowledge into action and then evaluating if these actions are working. In my latest Tough Things First podcast, I discuss why the habit of self assessment and reassessment is worth developing. https://lnkd.in/dCq5jc-2
the Dunning of the Dunning-Kruger Effect recently discussed the Dunning-Kruger Effect, (interviewed by Corey Powell, in the Open Mind Podcast) https://lnkd.in/ekFXnVDn
Chapter 28 covering the conclusions of the events on the Incident Ground from the Phase 1 Grenfell Tower Inquiry report are now available on the Fire Ninja podcast.
https://lnkd.in/epbZpdCA
