Good morning! This is your daily ☕️ Techpresso.
In today's Techpresso:
⛑️ Google scrambles to manually remove weird AI answers in search
🎮 Apple built a Tetris clone for the iPod but never released it
🚨 Deepfake Biden robocall creator indicted, faces $6M fine
🙃 EU users can opt out of Meta's AI data use... but they'll have to justify their request
🤔 New study says ChatGPT gives wrong programming answers 52 percent of the time
🎁 + 7 other news you might like
🔮 + 2 handpicked research papers and tools
⛑️ Google scrambles to manually remove weird AI answers in searchLINK
Google is facing issues with its new AI Overview product, which provides strange answers like suggesting users put glue on pizza and eat rocks, leading to a manual removal of these features.
Although Google has tested AI Overviews for a year and served over a billion queries, the rapid cost reduction might have been premature, resulting in low-quality output.
Google aims to improve its AI systems despite current challenges, but experts indicate that the final step to achieve 100 percent accuracy is extremely difficult and requires reasoning akin to human fact-checking.
🎮 Apple built a Tetris clone for the iPod but never released itLINK
Apple Demo found a prototype third-generation iPod containing a Tetris clone called Stacker, which was never released commercially.
The prototype also included other unreleased games like Block0 and Klondike, and had a "DVT" label on its back indicating it was in the middle stage of development.
Despite contacting former Apple executive Tony Fadell, the reason for Stacker's non-release remains unclear, though Apple later released officially licensed Tetris games on its "Classic" iPod models.
🚨 Deepfake Biden robocall creator indicted, faces $6M fineLINK
Steven Kramer has been indicted on felony charges of voter suppression and misdemeanor impersonation of a candidate for creating a deepfake anti-Biden robocall.
Kramer faces a $6 million fine from the FCC for using AI-generated voice cloning to impersonate President Joe Biden and employing caller ID spoofing tactics to hide the call's source.
Kramer hired a telemarketing firm to play the recording to over 5,000 voters, which aimed to suppress voter turnout in the New Hampshire Democratic primary, and this marks the first-ever enforcement action against spoofed deepfake robocalls in the US.
🙃 EU users can opt out of Meta's AI data use... but they'll have to justify their requestLINK
Meta is informing Facebook and Instagram users in the EU via in-app notifications about changes to its privacy policy effective June 26, allowing them to opt out of their data being used to train AI models if they provide a justification.
Meta cites a "legitimate interest" in using user data like posts, comments, and mentions, excluding private messages, to train its AI models, and users can refer to the GDPR when explaining why they want to opt out.
The EU authorities are expected to review Meta's claim of "legitimate interest" and the opt-out procedure's adequacy, while an interim report indicates ongoing concerns about data protection in AI, particularly with OpenAI's ChatGPT.
🤔 New study says ChatGPT gives wrong programming answers 52 percent of the timeLINK
A study from Purdue University found that ChatGPT answers computer programming questions incorrectly 52% of the time and is verbose 77% of the time.
The research, presented at the Computer-Human Interaction Conference, also revealed that users preferred ChatGPT's answers 35% of the time due to its comprehensive and well-articulated language.
Despite these preferences, the study highlighted that users overlooked incorrect information in ChatGPT’s responses 39% of the time, emphasizing the need for better accuracy and awareness of misinformation risks.
Other news you might like
Google Search’s “udm=14” trick lets you kill AI search for good.LINK
The 'Doge' dog has died.LINK
Microsoft CEO Satya Nadella says AI should not be called AI.LINK
TikTok will now let companies create social media campaigns using generative AI.LINK
From Plaid to Figma, here are the startups that are likely — or definitely — not having IPOs this year.LINK
PayPal Mafia’s David Sacks on his new AI-powered work chat app rivaling Slack.LINK
EU data watchdogs tell OpenAI there's more work to do on ChatGPT.LINK
Latest research and tools
Spot: a simple, cross-platform, reactive desktop GUI toolkit for Go that allows for easy creation of native applications with broad widget support and automatic UI updates.LINK
Mistral Fine-Tune: a lightweight codebase that supports memory-efficient and high-performance fine-tuning of Mistral models using LoRA, which involves freezing most model weights and training only a small percentage of additional low-rank matrix perturbation weights, optimized for A100 or H100 GPUs and designed for multi-GPU-single-node setups.LINK
Want to get the latest news differently? Find us on:
See you tomorrow for a new dose of ☕️ Techpresso!
-min.png){kind=small}
.png){kind=small}
