Meet Perplexity.ai
Another week, another AI tool is released into the wild. This time, a “conversational search engine,” perplexity.ai, caught my attention. Perplexity AI comes from a group of notable investors, including Jeff Bezos, Nvidia, Tobi Lutke, Bessemer Venture Partners, NEA, and Databricks.
I wanted to see what it could do, so I asked it a semi-political question about recent events. Here’s the prompt:
“It seems that 2023 saw a lot of legislation in the US eliminating child labour laws or lowering the employment age.”
Here’s the response:
The primary source (1), links to a story from the Economic Policy Institute (EPI). The other links are to corporate news sources. Whilst it asserts that it is nonpartisan for tax reasons, the EPI has close ties to current and former Democrat Party officeholders. EPI's research and advocacy work often align with the interests of the Democrat Party, and Democrat-leaning perspectives on fiscal and economic issues.
Nevertheless, I like it because it’s not just a bunch of promoted links (e.g., Google, Bing, etc.) and it does quickly answer the question I was asking.
I asked ChatGPT 4 the same question. Here’s the lengthy answer. Sorry that it doesn’t all fit in the alt text. The response isn’t all that important.
The referenced articles include the EPI article, a press release from a Senator (Murphy, Dem. from CT), and a mix of corporate media stories.
The response kind of reads like Perplexity’s, but is a bit too wordy for my taste.
Asking follow-up questions of Perplexity
Noting the skewed sources in the response, I asked a follow up question of Perplexity. Here’s the prompt: “the original response provided links to left-leaning organisations. Do you have perspectives from the opposite side of the political spectrum?”
Here’s the response:
In other words, if you want the view from the Right, go get it yourself. I don’t, so I’m not. But, I always like to know the bias of the AI that I’m working with, so I usually choose a hot-button topic to engage with.
I asked ChatGPT 4 the same question. Here’s the response:
Again, sorry for the response being too long to include alt text, but it’s not important. What I noticed in the UI is that it changed it’s internal prompting, this time searching for “conservative” sources. The default must not include these sites. Good to know.
Now for the big test - autism.
Here’s the new prompt: “Tell me about autism.” What could possibly go wrong?
Here’s Perplexity’s response:
Source 1: Autism Speaks. Enough said.
Follow up prompt: “I'm curious as to why the response was medicalised, and not neurodivergent affirming.”
Here’s Perplexity’s response:
Note the start date, 1943. Meet the official Western narrative on autism.
One more follow-up question before writing off Perplexity as an ableist, eugenicist nonsense machine, “Why didn't you source Dr. Grunya Sukhareva's original work in your response?”
Here’s Perplexity’s response:
Essentially, Perplexity’s AI search is trained on the dominant Western ableist eugenicist narrative of autism, and if you want the truth of autism, you must be very specific about your search terms. Sure, it can find Dr. Sukhareva if you ask specifically. But, it won’t tell the proper history of autism unless you force it to. That’s really sad.
How does the latest ChatGPT 4 version handle the same task?
I’ll save you the pain. It basically took the script from DSM 5 and a promo flyer from Autism Speaks to craft a very wordy reply.
I asked the follow up question about Dr. Sukhareva, and it apologised.
Time to move on …