Perplexity's election day AI

🇺🇸 Happy Election Day Folks!

Elections in the USA can take weeks to finalise. Forget one-night results; in the U.S., vote counting can take weeks thanks to mail-in ballots and different state rules. So if you’re waiting for quick answers, buckle up — Election Night’s more like a season finale stretched over multiple episodes.

What’s in store:

Read Time: 4 minutes

Perplexity has launched an Election Information Hub to see if AI can reliably provide important voting information.

And just in time for the hottest election to ever hot.

Announced on Friday, the hub includes AI-generated answers to voting questions, and candidate summaries, and will track live vote counts today, Election Day, using data from The Associated Press (AP).

Perplexity’s voter information, which covers polling requirements, times, and locations, comes from Democracy Works, which also supports Google’s similar features.

The election-related answers are drawn from a set of trusted, fact-checked sources like Ballotpedia and news outlets.

Perplexity spokesperson Sara Platnick confirmed that both AP and Democracy Works are official partners.

She added that Perplexity is monitoring its system to keep these sources prioritised when answering election questions.

The hub lets users enter a location (e.g. an address or city) to see what’s on their ballot.

There are also tabs for tracking the presidential, US Senate, and US House elections, which are state-by-state updates showing the percentage of votes counted and who’s leading.

Here’s the rundown:

However, there have been some issues.

Some summaries missed key details, such as not noting Robert F Kennedy’s withdrawal from the race.

Additionally, a listing for “Future Madam Potus” mistakenly linked to Vice President Kamala Harris’s summary and included meme images, which was later fixed.

These problems show the difficulty of using AI for high-stakes information which is why other AI companies like OpenAI, Meta and Microsoft choose to direct voters to established resources like canivote.org or Google Search.

Perplexity really said “trust me bro”.

Vote for live results and see results + opinions from yesterday at the bottom of the email.

Tired of low engagement in your outreach campaigns?

lemlist helps you create ultra-targeted, AI-powered multichannel campaigns in seconds, so you can:

November 5th 1935 - Parker Brothers begins marketing the board game Monopoly.

New York City tools out an AI chatbot to help residents navigate the ins and outs of starting and running a business. Sounds awesome right? But fast forward five months, and this helpful bot is giving advice that’s not just wrong—it’s illegal.

Testing by The Markup showed the bot confidently telling landlords they don’t need to accept Section 8 vouchers or rental assistance.

Spoiler alert: that’s a hard no.

In NYC, discriminating against tenants based on income source is against the law (with a tiny exception for some small buildings). One housing expert even called the bot’s advice “dangerously inaccurate.”

So while the AI chatbot might sound official and authoritative, its guidance could land you in hot water.

MIT shared a new method for training robots that moves away from traditional, smaller data assets.

Training wheels: OFF.

Inspired by LLMs, this approach uses large, mixed data like powering models such as GPT-4.

The goal is to help robots handle complex tasks better.

Robots often learn by copying human actions, which works well in controlled environments.

However, the method truffles when robots face new challenges, like different settings or lighting.

The limited range of training data means robots can’t easily adapt.

To solve this, MIT’s team developed Heterogeneous Pretrained Transformers (HPT).

This new model brings together data from various sensors and environments.

A transformer model is used to process and train the robot, and bigger transformers produce better results.

Users can enter specific robot designs, configurations, and tasks directly into the system.

Here’s what you should know about this:

The big vision is a “robot brain” that could be downloaded and used without extra training.

Though it’s early, the researchers, supported by the Toyota Research Institute, hope that scaling up this method will lead to major breakthroughs, similar to those seen with LLMs.

This builds on TRI’s other work, including joining research with Boston Dynamics’ hardware.

No more copying from homework anymore!

Space: Rocket Lab plans to challenge SpaceX in military space launches, confirming its bid for Pentagon contracts with its medium-lift rocket, Neutron.

Crypto: Blockchain-Ads, a leading network for crypto audiences, has introduced its Partner Program with LeanMakreting-Crypto as its first certified partner.

Games: Supercell has boosted its stake in Space Ape, set to fully acquire the London studio by spring 2025 and open its first UK office.

Sports: Liverpool's Ibrahima Konate confirmed his arm injury from the Brighton match isn’t serious and he’ll be fit for the team's next game.

Don’t Miss: X has begun rolling out a change allowing blocked users to still see posts, following and followers lists, which may raise harassment concerns. While blocked users remain unable to interact with posts, critics argue this reduces the blocking function’s effectiveness.

Submit your artwork to Mindstream →

Yes, with the right approach. 48%

Not sure, seems tricky. 52% ✅

Your Views:

“Legislation ALWAYS lags behind innovation. So far, attempts to regulate BEFORE innovation settles down, has created road blocks to the innovation!” - larry

“Must!. But not now, yesterday...” - wir.u

Submit your opinions in our daily poll to be featured!