- AI Tangle
- Posts
- ☕️ Yahoo's Grand Plans with Artifact
☕️ Yahoo's Grand Plans with Artifact
As the week's curtains closes, so do the blinds on Artifact, as Yahoo acquires the AI news app from Instagram's co-founders with some plans to capitalize on Artifact's underlying tech. Meanwhile, Anthropic discovers a perplexing jailbreak technique for LLMs, internet browser company Brave gifts its Leo AI assistant to Apple users, and Elon Musk continues his war against OpenAI. Join us at AI Tangle as we untangle this week's happenings in AI.
THE BIG AI STORY
A recent development in the world of AI sees Yahoo acquire the AI-powered news app Artifact, created by Instagram co-founders Mike Krieger and Kevin Systrom. Despite Krieger and Systrom announcing Artifact's death a few months prior, its unique approach to news caught the attention of Yahoo with plans to give its Yahoo News and other platforms an AI-infused boost.
So what will happen?
Both sides emphasize that Yahoo is after the underlying tech of Artifact rather than the talents behind it, though the two co-founders will be acting as "special advisors." Artifact itself will be destined to disappear, but its underlying technology, with a focus on the app's categorizing, curating, and personalized content features, is planned to be incorporated into Yahoo News, which has a user base of 185 million visits a month.
COLLABORATING WITH ABOUTHIRE
Want to save on your talent costs? AboutHire We helps great companies hire talent from Northern Europe. You can now hire amazing employees that cost 60% less than US equivalents. Learn more at www.abouthire.com
6 QUICK HITS
Recently, Anthropic researchers wrote a blog post detailing the discovery of a "many-shot jailbreaking" vulnerability, a method used to bypass the safeguards of an AI model, where it can be tricked into answering discriminatory, deceitful, or outright dangerous questions. The researchers believe that the issue arises from the increased "context windows" of LLMs. Despite mitigation efforts being underway, the vulnerability currently impacts the company's models and those of other firms.
AI posterchild OpenAI seems to be expanding its Custom Model program to help enterprises develop tailored generative AI models. The newdly-added components included in the program are "assisted fine-tuning" and "custom-trained models" to maximize performance and lessen the strain on OpenAI’s model serving infrastructure. OpenAI also recently unveiled model fine-tuning features for developers working with GPT-3.5, such as a dashboard and support for third-party integrations, but hasn't spilled the beans on anything about GPT-4.
Two supposed crypto scammers recently got an angry letter in their inbox as Google sues them for using its Play Store app store to upload fraudulent cryptocurrency and financial investment apps to swindle users out of their funds. The two app developers, one based in Hong Kong and the other in China, allegedly uploaded 87 different apps and got around 100,000 downloads combined starting in 2019. In the court, Google is seeking an unknown amount of compensation for damages and to block the developers from committing further fraud.
Brave, a popular open-source internet browser, has launched its privacy-centered Leo AI assistant following its Android and desktop release. Leo provides voice-to-text capability, unique to the iOS version, and can perform tasks like summarizing pages, generating reports, and writing code. As of current, Leo uses Mixtral 8x7B as the default but offers alternative models such as Claude Instant and LLaMa 2 13B, as well as other models and higher rates with its Leo Premium subscription.
As if Elon Musk's relationship with OpenAI couldn't get more dire following the lawsuit between him and the company, according to Elon Musk himself, he is currently reportedly in a talent war to prevent Tesla's engineers from defecting to Sam Altman's OpenAI instead of his xAI startup, which he aims to achieve by increasing compensation. Some investors worry that xAI could use the talent more than OpenAI, as the company's premiere chatbot, Grok, is not seen as a legitimate threat to the likes of GPT-4.
SiftHub, an AI startup founded by Manisha Raisinghani, former CTO and co-founder of LogiNext, has raised $5.5 million in seed funding, with plans to flesh out its AI assistant more. SiftHub's AI assistant aims to boost sales efficiency by automating non-sales tasks like data entry, filing requests, researching customer info, and other similar duties. At the moment, the company's assistant is available through Slack, Microsoft Teams as an add-on, a Chrome plugin, and a web app, with support for over ten languages.
4 AI TOOLS
Decktopus - Create polished presentations from case studies to sales decks with ease using Decktopus, the AI-driven presentation tool built for use by both individuals and organizations alike.*
Flowsage - Efficiently design, collaborate, and create flowcharts with Flowsage, an AI-powered tool for seamless idea visualization.
Senuto - Senuto covers a business's SEO needs with its expansive suite of data and content tools to help recognize your strengths and weaknesses in SEO.
Typper - Elevate your business intelligence processes with Typper, an AI-powered BI tool designed for custom data integration to ensure a seamless analysis experience.
* indicates a sponsored tool
AI READ & WATCH
How Do You Actually Safety-Test AI? (10-min read)
Explore the challenges of safety-testing AI systems in this TIME article, discussing the efforts of Beth Barnes' METR and AI giants on developing safety tests for advanced AI systems and whether it's even feasible with current technology.
The Need For Stargate (19-min watch)
AI Explained, an in-depth AI content channel, is joined by Perplexity CEO Aravind Srinivas to break down OpenAI's plea to Microsoft to build Stargate, the $100 billion supercomputer.