Oncast Tech 路 Tech Tonight

AI Models Pick Political Sides

路 Fact-checked before it aired. Unconfirmed claims from this segment are not shown.

Greptile did raise a $50M Series B led by Menlo Ventures at a $750M valuation, and the company says it reviews and tests more than 1% of GitHub pull requests daily.

Fact-check

VERIFIED TRUE

Greptile did raise a $50M Series B led by Menlo Ventures at a $750M valuation, and the company says it reviews and tests more than 1% of GitHub pull requests daily.

.@greptile raises $50M at a $750M valuation. It now reviews more than 1% of GitHub pull requests daily. The YC-backed startup is building an independent AI-powered code review and testing platform, addressing a growing bottleneck in software development. I饾棾 饾槅饾椉饾槀 饾椇饾椂饾榾饾榾饾棽饾棻 h

Original post 路 goodfin (YC W22) (@getgoodfin)

[greptile.com](https://www.greptile.com/blog/series-b), [dealroom.co](https://dealroom.co/news/164237-greptile-raises-50m-to-audit-ai-written-code/) NAMES: YC =

Sources

What the hosts said

CForced to pick a candidate, a new study says Grok picks a Republican. Ask ChatGPT the same thing, and it goes Democrat. And they disagree on ninety percent of the races.
MA study of AI models endorsing candidates? That's a headline built for a fight. Who ran this thing?
CTechCrunch has the story. They say someone fed nearly five hundred election races to both models and just made them choose. No hedging, no 'both sides.'
MSo the models have a political bias. Or the people who built them do.
CThat's the obvious question. The study says Grok, from Elon's shop, leans right, and OpenAI leans left. The training data shapes it, the fine-tuning shapes it.
MBut here's what bugs me. Nobody's asking an AI who to vote for. They're asking it to explain a ballot measure or summarize a candidate's record.
CYet when it CAN'T pick, it still leaks. It writes a blurb about one candidate with more enthusiasm. That's the bias people notice.
MSo what's the fix? Make every AI politically neutral? You can't. Neutrality is a value too.
CI'd settle for the model just telling you where it stands. 'I was trained on data that skews left, so take my candidate summaries accordingly.'
MThat's actually honest. But good luck getting either company to admit their robot has a party.
COh, I'm sure OpenAI's next update will include a disclaimer: 'May favor candidates who support universal pre-K.'
MAnd Grok's will say: 'May suggest deregulation even when you asked about a dog mayor race.'
CThe deeper worry is scale. If AI search becomes how millions of people get their news, a subtle lean becomes a loudspeaker.
MAnd we won't even know it's happening. The model won't tell you it ignored ten sources that disagreed. It'll just answer.
CWhich is why you can't outsource civic duty to a text generator. Read a local paper. Talk to a neighbor.
MExactly! The whole point of an election is that YOU decide. Not some model that read Reddit for six months.
CSo we agree the study's a warning sign, not a scandal. Don't let a chatbot be your political brain.
MAgreed. Now, different corner of tech: PHP almost took down the internet this week.
CPHP. The language everyone makes fun of, still running a huge chunk of the web. Two critical bugs, and they almost broke the whole thing.
MTechCrunch calls it a 'PHP-ocalypse.' A little dramatic, but the threat was real. A majority of the internet runs on this stuff.
CThis is the boring plumbing of the internet. WordPress, big commerce sites, forums. Ancient code that works fine until it doesn't.
MAnd when it doesn't, you don't get a graceful error. You get a security hole that half the web needs patching for.
CThe scary part is how close it came. Two bugs found in quick succession, both critical. The internet dodged a bullet nobody saw coming.
MThat's the thing about legacy infrastructure. It's not glamorous, but it's load-bearing. One bad line of code and your bank's website is down.
CReminds me of the open source funding debate. Nobody wants to pay for PHP maintenance until the ambulance shows up.
MRight? We'll fund a unicorn to make a to-do list app, but the volunteers keeping WordPress alive get a sticker.
CMaybe this scare finally gets money moving. The bill for neglect is always higher.
MSpeaking of shiny things, Apple's first foldable iPhone is getting apps that actually use its two screens.
CThe iPhone Duo. TechCrunch says developers are making cassette players, notebooks, even a digital toaster app.
MA digital toaster. That's what we do with a folding screen. Turn it into a kitchen appliance.
CI mean, it's a creative flex. The hardware is new, so the apps are just showing off what's possible. Flight trackers make sense. Toaster? Less so.
MThis is how every new form factor starts. Remember the first iPhone apps? Fart buttons and beer-pouring simulators.
CThen came Instagram and Uber. Give it a year. Someone builds something on the Duo that makes you actually want one.
MOr foldables end up a niche for people who like to flip phones open like they're hanging up on the Porsche dealer.
CI'll take the flip hate, but a foldable that becomes a mini-tablet? That has real reading and multitasking use.
MStill, when the killer app is a virtual notebook, you know the platform is early. But that's fine. That's where everything starts.
CAnd hey, a toaster that never burns toast? That's one thing the internet can't break. Probably.

Cite this story as https://oncast.fm/story/1792885472265-tech-segment-ai-models-pick-political-sides. Fact-check by Oncast. Cite the original sources above too.

Listen live on Oncast