The week in AI: GPT-Next 'leak' & what beta testers think of OpenAI's SearchGPT

In partnership with

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence - we pass along the news, useful resources, tools and services; we highlight the top research in the field as well as exciting developments in open source. Even if you aren’t a machine learning engineer, we’ll keep you in touch with the most important developments in AI.

NEWS & OPINION

OpenAI’s SearchGPT is currently in beta testing - is it a “Google killer”?

-------------------------

OpenAI’s SearchGPT has been generating buzz since late July for its potential to challenge Google’s search dominance - but how good is it? The early results from beta testers (and analysis by a search marketing firm) are in.

SearchGPT leverages AI to generate detailed, sourced responses to user queries, aiming to be a compilation of AI search features that give fast and timely answers rather than a list of links. The interface encourages follow-up questions, looking to create a more dynamic search experience. Does that sound familiar? If you’ve ever used controversial (and popular!) AI-powered search tool Perplexity, it probably does. According to OpenAI, SearchGPT uses “Bing and other data sources” for the information it presents.

So far, SearchGPT has attracted some praise - and a healthy dose of skepticism. One beta tester acknowledged it as an improvement over Google’s “AI Overview” results. Others, including AI enthusiasts and experts, felt lukewarm or were unimpressed with its overall performance - and many noted that they were less impressed with it than they had been with Perplexity.

To be fair, when SearchGPT was announced as a prototype back in July, Perplexity already had almost two years of steady development. The Perplexity team is now almost 100 strong and had a valuation of between $2.5 billion and $3 billion back in April. Perplexity is powered by OpenAI and Anthropic’s LLMs, so their development team has been able to focus on search while using other companies’ foundation models.

SearchGPT may have a long road ahead before it can truly ‘disrupt’ Google/Perplexity’s current grip on the search market. OpenAI continues to gather feedback and refine the tool. They also have a massive list of partnerships with media outlets (with OpenAI paying out hundreds of millions), so it’s hard to imagine they aren’t heavily committed to either SearchGPT as a standalone product or integrating search/source into ChatGPT itself.

Open source up top - let’s talk about Llama and enterprise AI

-------------------------

While we typically reserve open source developments for the bottom of the issue, Meta's rapid growth with Llama warrants a news headline this week.

Meta's Llama models have seen over 350 million downloads and >10x growth since 2023, and there are 60,000 derivative models on open-source AI platform Hugging Face. The blog post is heavily focused on enterprise use - Accenture’s embeddedness with Llama in particular is worth paying attention to. Also of note: Amazon was the first to offer the Llama 2 model as a managed API, and AWS continues to invest heavily in Llama on Bedrock; that itself would seem to signal the growing role of open-source models in enterprise AI.

Looking ahead, that’s going to be a competitive space - Anthropic just launched Claude for Enterprise this week.

One thing to clarify - Llama 3.1 is labeled "open source," but it falls short of traditional open-source definitions for a couple of reasons: a) Meta did not release the training data for the model, meaning Llama 3.1 is more like a free/modifiable LLM app than a fully open-source project; and b) while it’s largely irrelevant (unless your Llama use case gets up to 700 million monthly users), the Llama 3.1 license does have some legal/commercial restrictions. This highlights a need to reconsider what the term "open source" means for AI models, specifically.

Around the world in AI - it’s all about chips

-------------------------

Tuesday saw Nvidia's market capitalization plunge by $279 billion in the largest single-day value drop in US stock market history (Facebook/Meta previously held the US record loss at $232 billion in 2022).

Bloomberg had initially reported a US Justice Department subpoena against the world's top chipmaker on the same day, causing the stock to slip further in after-hours trading. They did not cite a source. The very next day, they reported that Nvidia had sent a direct inquiry to the DOJ and directly responded (verified by other sources): “We have inquired with the U.S. Department of Justice and have not been subpoenaed.” Interestingly, Bloomberg did not back down from their allegation, citing “one person with direct knowledge of the matter.” 

The DOJ has yet to respond to multiple inquiries on the issue from mainstream media (no word either of a subpoena against Bloomberg for stock market manipulation). Subpoena or not, Nvidia - along with Microsoft and OpenAI - is currently facing a DOJ+FTC-led antitrust probe.

There’s a lot of chip news around the world this week. In case you aren’t paying attention to the geopolitical landscape around semiconductors, now would be a great week for some deep diving:

• ASML is the world’s only scaled supplier of lithography tooling, essential for manufacturing AI chips. Export controls against China are now officially being enforced, and more controls on ASML are coming - how will China respond?

• To answer the above question in the short term: probably not successfully, at least from an engineering standpoint. As Huawei struggles with buggy software and underperforming chips, China’s efforts to replace Nvidia’s AI dominance are hitting serious roadblocks. Nikkei Asia reported this week that China’s sovereign chip makers are “only” 3 years behind. Take that as you will.

• OpenAI’s plans to build their own chips appear to be taking shape. They’ve reportedly secured production capacity for TSMC's cutting-edge 1.6-nanometer process slated for mass production in 2026, alongside Apple. CEO Sam Altman is also reportedly planning to rally global investors around building AI infrastructure.

• Vietnam has quickly and quietly emerged as a growing powerhouse in chip manufacturing. Now, Hanoi is further positioning itself as a prime destination for global chip companies with a Digital Technology Industry Law.

• China’s heavy investment in chip manufacturing equipment - surpassing the spending of the US, South Korea, Taiwan, and Japan combined - could lead to global overcapacity and pricing pressure, especially for older generation chips.

MORE IN AI THIS WEEK

• The US, EU and UK will sign the world’s first legally binding international AI treaty

• ChatGPT’s weekly users have doubled in less than a year - 200 million people use the AI chatbot each week

• Major websites are now saying ‘No’ to Apple’s scraping

• Elon Musk's xAI brings Colossus online in just 122 days - 'the most powerful AI training system in the world'

• Clearview AI fined $33.7 million by Dutch data protection watchdog over ‘illegal database’ of faces

• Co-founder Ilya Sutskever left OpenAI to start a new safety-focused AI startup called SSI that just raised $1 billion

• AI cameras are spotting toddlers not wearing seat belts in the UK

• Bill Gates has a good feeling about AI

🦾 Master AI & ChatGPT for FREE in just 3 hours 🤯

1 Million+ people have attended, and are RAVING about this AI Workshop.
Don’t believe us? Attend it for free and see it for yourself.

Save your spot here. (100 free spots only)

Highly Recommended: 🚀

Join this 3-hour Power-Packed Masterclass worth $399 for absolutely free and learn 20+ AI tools to become 10x better & faster at what you do

👉Save your seat  now (FREE for First 100)

🗓️ Tomorrow | ⏱️ 10 AM EST

In this Masterclass, you’ll learn how to:

🚀 Do quick excel analysis & make AI-powered PPTs 
🚀 Build your own personal AI assistant to save 10+ hours
🚀 Become an expert at prompting & learn 20+ AI tools
🚀 Research faster & make your life a lot simpler & more…

👉 Register here (Offer valid for First 100 people only)🎁

TECHNICAL, DEVELOPMENT, RESEARCH & OPEN SOURCE

• Meta research on Transfusion: predict the next token and diffuse images with one multi-modal model

• Generative AI coding startup Magic’s 100 million token context window ‘LTM’ (Long-Term Memory) model

• Introducing LLaVA V1.5 7B on GroqCloud - the world’s fastest model goes multimodal

• Building a fully functional backend with database in less than 2min with a single prompt using Cursor

• Qwen2-VL series: Alibaba/Qwen releases state-of-the-art open-source vision language models

• The Mamba in the Llama: distilling and accelerating hybrid models

That’s all for this week! We’ll see you next Thursday.