In-Depth Leak Reveals Exclusive Information on GPT-4, the Fourth Generation of OpenAI’s Powerful Language Model


GPT-4 details are leaked

An expensive write up on most of the secret details behind GPT-4 was leaked. How it was built, how much the training cost as well as most other details that we’ve been speculating about since it’s release.

One interesting reveal is that GPT-4 is not one model like we’ve been led to believe. As well as a look into why gpt-4 seems to be getting dumber (or “getting a lobotomy” as some put it).

Is this the beginning of the end for OpenAI? Will we see an open-source GPT-4 model hit this year?


Anthropic releases Claude 2, its second-gen AI chatbot

Anthropic, an AI startup, has unveiled Claude 2, their second-generation AI chatbot, now available in beta in the U.S. and U.K. This new and improved version outperforms its predecessor in several areas, like tackling the bar exam, coding in Python, or solving math problems. It can also format outputs in JSON, XML, YAML, and markdown.

Claude 2 uses “constitutional AI,” a technique developed by Anthropic, guiding the model’s behavior through set principles. The aim is to make it easier to tweak and adjust the AI’s behavior as needed. But, the company admits as the AI gets smarter, it gets trickier to predict its behavior.

The company’s ultimate goal is to create a self-teaching AI algorithm that could revolutionize virtual assistants and the way they perform tasks, like answering emails or generating art. Claude 2 is a stepping stone towards this goal, but there’s still a long way to go.

Anthropic is in competition with OpenAI, Cohere, AI21 Labs, and others in the AI landscape. Despite being a new kid on the block, it has already raised significant capital but estimates needing $5 billion more over the next two years to fully realize their ambitious chatbot vision.


eBay acquires AI-powered product authentication company Certilogo

eBays been on a shopping spree, but it ain’t for knick-knacks or collectibles. They’ve just snagged Certilogo, a tech company from Italy that uses fancy AI to make sure the clothes and fashion items you’re buying aren’t knock-offs. The folks from Certilogo will keep doing their thing, under the watchful eye of their big boss, Michele Casucci.

Certilogo’s tech helps brands keep tabs on their products from the factory to your closet, and gives shoppers a way to double-check that their new threads are the real deal. This buyout shows that eBay’s keen on making sure used fashion sold on their site is legit.

eBay’s VP, Charis Marquez, sounds stoked about the new deal. She says this’ll help brands keep their products safe from counterfeiters and will give buyers peace of mind knowing they’re not getting ripped off with fakes.

Back in May, eBay hinted that they were planning to invest in the secondhand fashion market. They reckon that snagging Certilogo will make them a go-to spot for folks looking for used clothes. They also think it’ll help buyers feel good about making more eco-friendly purchases.


Microsoft rolls out AI-generated ad headlines and descriptions

Microsoft’s rolling out a new tool that writes ad headlines and descriptions using artificial intelligence (AI). This tool will make it easier for marketers to come up with catchy headlines without needing to scratch their heads for hours. It’ll also suggest ways to improve your ads and save you time.

Here’s how it works. The AI tool is available in 35 languages and it can figure out what language to use based on your website. You find the tool under the “Create an Ad” tab. Once there, you type in your final web address (URL) and the tool will offer several headline ideas based on that URL. You can then pick the best option with a single click.

Microsoft is also launching two more features to help advertisers. The first, auto-generated assets, helps reduce the work of creating separate ad campaigns for different devices and audiences. The second, called IF functions, lets you change the ad based on the device or audience. These two features are expected to roll out over the next month.


Amazon charts its own course on A.I., leaning on its strengths in cloud and e-commerce

Amazon is zigging while others zag, doubling down on AI tools for developers and businesses rather than going all in on customer-facing tech like chatbots. They’re using their heavy-hitter, Amazon Web Services (AWS), to make it happen. 

Don’t get it twisted, Amazon is also investing a cool $100 million in an AI innovation center for free training. Plus, they’ve got AI code generators like CodeWhisperer to help developers speed up their work. And while the stock market’s been all about Nvidia, Microsoft, and Alphabet this year, some analysts think Amazon is just quietly flexing for the next big wave in AI.

Keep in mind, AWS isn’t just an AI playground – it’s also a dominant player in the cloud storage and computing market. And even though growth has slowed a bit due to competition and cost-cutting, it’s still a cash cow. Meanwhile, Amazon’s also using AI to power its e-commerce and advertising game, making searches and ads more relevant to users.


Shutterstock expands deal with OpenAI to build generative AI tools

Shutterstock is doubling down on its partnership with OpenAI to make AI better and smarter. OpenAI will use Shutterstock’s pictures, videos, and music to train its AI, and in return, Shutterstock will get early access to OpenAI’s latest gadgets and new ways to jazz up images in its library. They’re also going to bring this AI tech to folks on their phones through Giphy, a GIF library Shutterstock recently scooped up.

Shutterstock’s move might seem odd, considering the ongoing tug-of-war between stock image galleries and AI startups. These AI companies can whip up customized stock images like hotcakes, which makes some folks in the image gallery biz pretty nervous. In fact, there were heated protests from artists and photographers who believe AI startups are making a buck off their work without giving credit where it’s due.

Unlike Getty Images, Shutterstock decided not to risk a long-winded court fight. Instead, they’ve buddied up with OpenAI to create a picture-making AI. Shutterstock has also cut deals with other big names like Nvidia, Meta, and LG to make even more AI tools.

Trying to keep the peace with artists, Shutterstock has set up a “contribitor fund” to pay artists for using their work to train their AI. Artists also get a cut of the pie whenever their work is used to make new content.


Dashworks launches AI assistant to streamline internal knowledge for enterprises, raises $5M

AI startup Dashworks, straight outta San Francisco, just raised another cool $5M, bumping their total funding up to $9M. They’re gearing up to make waves with Dash AI, their smart assistant, which is like a turbo-charged filing cabinet for big companies.

Why? ‘Cause right now, workplace knowledge is scattered from here to Sunday, across apps like Slack, Google, Asana, and Notion. It’s a hot mess and a royal pain for employees trying to track down specific info.

Dashworks’ head honcho, Prasad Kawthekar, said their goal is to help teams reach their full potential by bringing some order to this chaos. Dash AI is their answer – an assistant that can find, summarize, and understand all that scattered data in real time. From drafting emails to debugging code, it’s got your back.

Their customers, including Swiggy, are loving it so far. The company’s also planning to keep making the assistant smarter and more helpful by integrating more apps and improving its ability to answer questions.


Prompt engineering startup raises $5M as demand for generative AI services scales, an AI startup, has bagged $5 million in seed funding from multiple investors including Y Combinator, Rebel Fund, and Eastlink Capital. The company’s getting attention for improving generative AI, essentially teaching computers to write content that sounds human. Given the rapid progress of AI and how many businesses could use these skills, folks are keen on what Vellum’s got cooking.

The founders, all former employees of another startup Dover, saw a huge amount of work in fine-tuning AI applications for tasks like writing job descriptions and emails. They thought there had to be a better way, so they created Vellum to develop tools that make AI work better and faster.

Generative AI, like the kind Vellum specializes in, means you can get an AI model to spit out results based on normal, everyday language. Basically, it can understand your question and give you a pretty good answer. This doesn’t just change AI, it opens it up to a whole new audience – from product managers to software engineers.

Vellum’s got tools to help companies check AI output, add context to prompts, and make sure their AI’s saying the right stuff. It’s not just about getting an AI system up and running – it’s about refining it to handle all sorts of curveballs.


KPMG Plans $2 Billion Investment in AI and Cloud Services

KPMG is laying down $2 billion for AI and cloud services through a beefed-up partnership with Microsoft, a sign that big accounting firms are pushing hard into tech. This move is set to rake in over $12 billion in revenue across five years, about 7% of KPMG’s total annual income. KPMG’s big plan? To streamline its tax, audit, and consulting services, giving its team the chance to focus more on strategic advice.

KPMG’s boss, Bill Thomas, is clear that this tech move isn’t about cutting jobs. Instead, it’s about beefing up the workforce with AI know-how. KPMG also gets an early peek at Microsoft’s AI assistant, Microsoft 365 Copilot, as part of the deal.

A chunk of KPMG’s investment will go to generative AI, a tech that businesses are keen to use for better financial management. It’ll help KPMG tackle environmental, social, and governance tasks by wrangling big data sets for tax reporting and speeding up audits.

This bold move comes at a time when growth in consulting businesses has taken a hit. But KPMG isn’t new to AI – they’ve been using it to help clients for a decade. Other big-name firms have also poured billions into tech to boost their services, and like KPMG, they’re also cozying up to Microsoft for its AI offerings.


AI tools are designing entirely new proteins that could transform medicine

Smart cookies at the University of Washington are using an AI tool, RFdiffusion, to make proteins in ways Mother Nature never did. These ain’t your typical proteins – they could revolutionize medicine, with big potential in stuff like vaccines and drugs.

The AI’s way of work is pretty much like how a program creates realistic pictures. This approach helps the AI shape proteins quickly and have ’em snugly bind to other biomolecules. And guess what? They’re not just cool on paper – they deliver in real-life tests too.

In just a year, these AI tools have totally shaken up the protein-making scene. Now, scientists can design proteins that perform just right. It’s like tailor-making proteins for specific problems.

Making proteins wasn’t always a breeze. It used to be like putting together a jigsaw puzzle in the dark. But AI tools like AlphaFold came in and changed the game, predicting how proteins look with spot-on accuracy.

But there’s room to get even better. Some techniques can only handle small proteins or can’t make diverse ones. That’s where RFdiffusion comes in. It whips up a random mix of protein building blocks (amino acids) and then tidies it up to produce realistic, but brand-new proteins.