- AI Odyssey
- Posts
- AI agents 🤖: hype & challenges....
AI agents 🤖: hype & challenges....
Hello there! 👋 Welcome back to AI Creator. Get ready for a dose of exciting AI updates and intriguing reads. Just scroll down to dive in!
⏰ Quick roundup:
Claire Merchlinsky
AI 'agents' return with HyperWrite's complex task assistant, yet face privacy and execution challenges, sparking debates on specialized agents.
OpenAI introduces enhancements to ChatGPT, featuring prompt examples, suggested replies, and GPT-4 default for Plus users.
Edinburgh Fringe features AI comedy, highlighting potential and limits, as comedians underline the irreplaceable human touch in nuanced humor.
In a future, homeowners repaint houses to influence AI-based property valuations, exposing the benefits and dangers of algorithmic real estate assessment.
Meta turns to AI for ad business revival, prioritizing engagement and monetization through AI-driven tools and agent personas.
🚀 Latest developments:
IBM offers free AI courses for developers.
D&D fans turned off by AI-generated art spurred a Hasbro unit into banning it.
Alibaba open-sources AI model to take on Meta and empower developers.
Microsoft kills Cortana in Windows as it focuses on next-gen AI.
Sponsored by MarketMuse
🚀 Elevate your content strategy with MarketMuse!
Create high-quality content for the whole customer journey faster. Our AI analyzes your entire website to give you personalized, objective insights so you can win in search. Trusted by thousands of content strategy and marketing teams. Book a demo.
💬 Longer reads:
Hype about AI ‘agents’ rises again
The buzz around AI 'agents' is making a comeback, driven by the desire for personal AI-powered assistants. HyperWrite, a new startup, has launched an AI-powered assistant that aims to tackle multi-step tasks like flight bookings and online shopping. The concept of AI agents gained momentum with the release of open-source agents like BabyAGI and AutoGPT, which could automate tasks such as market research and podcast outlines. However, the initial excitement waned as these agents showed limitations, often getting stuck in action loops. HyperWrite's CEO, Matt Shumer, shared that their assistant is built differently, utilizing a unique model focused on web navigation. The startup aims to develop its own foundation model, 'Agent-1,' to power the assistant.
Despite its promise, HyperWrite's assistant faced challenges during testing, occasionally being overly autonomous or slow in execution. Privacy concerns also arose, as the assistant collects user data for tasks that involve sensitive information. Shumer hinted at a potential future training approach involving users' web activities. Although the startup envisions navigating the entire web, some experts, like CRV principal Vivian Cheng, believe specialized agents for specific tasks, like food delivery, could be more practical. Notably, the assistant demonstrated ethical refusal when asked about sensitive content. (Go deeper)
_____
OpenAI introduces enhancements to ChatGPT, featuring prompt examples, suggested replies, and GPT-4 default for Plus users…
We’re rolling out a bunch of small updates to improve the ChatGPT experience. Shipping over the next week:
1. Prompt examples: A blank page can be intimidating. At the beginning of a new chat, you’ll now see examples to help you get started.
2. Suggested replies: Go deeper with… twitter.com/i/web/status/1…— OpenAI (@OpenAI)
5:50 PM • Aug 3, 2023
____
Is this the future of comedy? The AI acts taking to the stage at the Edinburgh Fringe
Artificial intelligence takes center stage at the Edinburgh Fringe, with comedian Courtney Pauroso's creation, Vanessa 5000, playfully poking fun at emerging technology. Pauroso remarks on the blend of fear and enjoyment in exploring AI's potential impact. Meanwhile, the show "Artificial Intelligence Improvisation" tests whether algorithms can deliver humor, demonstrating that while robots can be funny within specific contexts, creating nuanced comedy remains a human skill. Comedian Pierre Novellie suggests that the complexity of cultural references and human connection makes comedy a tough challenge for AI. (Go deeper)
____
Welcome to a world where AI can value your home
In a dystopian future, homeowners repaint their houses in specific colors to manipulate AI algorithms used by platforms like Google Street View for appraisal. AI has reached a point where it can accurately assess home values by analyzing images from sources like Street View. While this technology offers efficiency, it may miss crucial real-world details that human agents catch during physical visits. The rise of algorithmic real estate brings advantages like democratizing appraisals but also risks, including bias and the potential for homeowners to manipulate their properties for better valuations. (Go deeper)
____
Meta's big AI play: Shoring up its ad business
Meta is leveraging AI to revive its ad business and counter a revenue decline. Despite attention on Meta's open-source AI models, AI's main role is enhancing its advertising core. Executives are shifting focus from the distant metaverse to using AI for engagement and monetization, as highlighted in recent earnings calls. The company's AI-driven tools have contributed to revenue growth, with plans to roll out AI agent personas to enhance user engagement and advertiser interaction. (Go deeper)
🧰 Tool discovery:
OpusClip: turns long videos into high-quality viral clips for you to share on TikTok, YouTube Shorts, and Reels to increase social media reach.*
YouLearn: learn any YouTube video with AI.
Sana: AI-powered learning platform helps pioneering companies share knowledge, learn fast, automate admin and perform 10x faster.
💸 Funding news:
AI-driven decision-making platform, One Model, secures $41m funding from Riverwood Capital.
AI cybersecurity startup, Jericho Security, raises $3m in pre-seed funding from Era, Lux Capital, FoundersXFund, and others.
AI-driven drug formulation startup, Persist AI, raises $4m in oversubscribed funding round.
📚 More reads:
Local governments are using AI to send kids to magnet school or set bail for convicts—and they’re not even trying to regulate it. (Fortune)
Why ChatGPT is getting dumber at basic math. (WSJ)
‘Not everyone is trying to build god’: Runway CEO Cristóbal Valenzuela tries to dampen the doomerism. (The Information)
A Zoom call, fake names and an AI presentation gone awry. (NYTIMES)
Will AI be an economic blessing or curse? History offers clues. (Reuters)