Amazon Reportedly Training AI With Twice as Many Parameters as GPT-4

Generally speaking, models with small context windows tend to “forget” the content of even very recent conversations, leading them to veer off topic. GPT-3 was launched in November 2020 and boasted what was then an unheard-of number of 175 billion parameters (analogous to the synapses in a human brain). GPT-4 was released on 14 March 2023, and the number of parameters has not been disclosed. OpenAI has been criticised for reversing its policy of publishing as much as possible about its systems. It replies, not unreasonably, that if these models can cause harm in the wrong hands, it would be silly to make it easier for the bad guys to replicate them.

But it did a great job on something it wasn’t really trained specifically for, which speaks to the possibility of GPT-4 being much more friendly to non-English speakers. It knows what the dress looks like, but it might not know if it’s the right outfit for your interview. GPT-4, however, can be given images and it will process them to find relevant information. You could simply ask it to describe what’s in a picture, of course, but more importantly its understanding goes beyond that.

GPT-4 can see and understand images

Moonshot’s leaner approach suggests the industry’s supposed “scale or die” narrative isn’t as bulletproof as it looked. Put your brand in front of 10,000+ tech and VC leaders across all three days of Disrupt 2025. Amplify your reach, spark real connections, and lead the innovation charge. Starting January 4, 2024, certain older OpenAI models — specifically GPT-3 and its derivatives — will no longer be available, and will be replaced with new “base GPT-3” models that one would presume are more compute efficient.

Workers across the globe are finding new, inventive ways to use ChatGPT every day. However, using such a powerful tool to cut down on the time you’re spending on tasks comes with a variety of different considerations. ChatGPT’s new Assistants API is built on the same technology as the new custom GPTs, with the goal of “helping people build agent-like experiences within their own applications”. Upgrade your lifestyleDigital Trends helps readers keep tabs on the fast-paced world of tech with all the latest news, fun product reviews, insightful editorials, and one-of-a-kind sneak peeks.

Lovable becomes a unicorn with $200M Series A just 8 months after launch

At the time of writing, GPT-4 is trained on data that was collected up until August 2022, so it has no knowledge beyond that date. That creates severe limitations on what the AI can do, and means that as time goes on, it becomes less accurate due to lacking the most up-to-date information. Explore the future of AI on August 5 in San Francisco—join Block, GSK, and SAP at Autonomous Workforces to discover how enterprises are scaling multi-agent systems with real-world results. Even as early users found GPT-4.5 workable — albeit a bit lazy — they questioned its release. Join leaders from Block, GSK, and SAP for an exclusive look at how autonomous agents are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.

• OpenAI says the long context comprehension, paired with improvements in instruction-following, make the GPT-4.1 models “more effective” at powering AI agents, the latest frontier in AI.
• GPT-3.5 Turbo tokens are also 3x cheaper than they were for the previous version of GPT-3.5 with the 16K context window at $0.001, while output tokens are also half price, costing just $0.002 per token.
• It breaks ground in acknowledging the enormous resources marshaled to make the program operate.
• That’s a very approximate description of how the attention mechanism and token count work, but the general idea is of expanded memory and the capabilities that accompany it.
• It did better on LiveCodeBench, a coding benchmark designed to mimic real-world scenarios, scoring 53.7 per cent compared to DeepSeek-V3’s 46.9 and GPT-4.1’s lacklustre 44.7.

ChatGPT was good at acting like a human, but put it under stress, and you could often see the cracks and the seams. In fact, it can perform so well on tests for humans that GPT-4 was able to pass the Uniform bar exam in the 90th percentile of test takers. OpenAI also assures us that GPT-4 will be much harder to trick, won’t spit out falsehoods as often, and is more likely to turn down inappropriate requests or queries that could see it generate harmful responses. You’ll need to pay to use the new version though, as for now, it’s locked behind the ChatGPT Plus subscription.

At the DevDay conference, OpenAI employees built their own chatbot agents – and it looks like the sort of thing that any knowledge worker could do. GPT-4 Turbo can accept images as inputs as well as text-to-speech prompts. However, the drop-down menu that ChatGPT Plus has been using to switch between other OpenAI apps like DALLE-3, is being retired. Now, ChatGPT will work out what sort of output you need based on your prompts. ChatGPT has famously struggled to give accurate answers on events that happened after its training data set was cut off, which was initially September 2021, but this was extended to January 2022.

Although it requires no coding, you’ll need a basic level of technical knowledge to use this tool effectively. You can augment your assistant with information and data from your organization, although OpenAI reiterates that the data you input into the models will not be used to train them and that developers can delete the data whenever they choose. GPT-4 Turbo also has an enlarged 128K context window, which helps it take prompts equivalent to around 300 pages of text. In short, GPT-4 Turbo vs GPT-4 is a straightforward win for the newer model, but there’s so much more to it than that.

Better document processing

For each of the below changes/announcements, we’ve provided information on which account holders can access the different language models. OpenAI held its annual DevDay conferenceand used it as an opportunity to announce a raft of changes to ChatGPT and other products, including wholesale price reductions for developers and a brand new language model for the chatbot called Turbo. Here’s what it is and how the key GPT-4 Turbo vs GPT-4 differences you should about. GPT-4 is a multimodal language model AI, which means it can understand text and other media, like images. This might sound familiar if you’re had a go with Stable Diffusion AI art generation, but it’s more capable than that, as it can respond to images and queries.

Simply put, AI agents are AI systems that can do tasks for you independently without being instructed on how to carry out every individual step. OpenAI says the GPT-4.1 models were built using developer feedback to improve areas they are particularly focused on, such as following reliable formats, adhering to response structure and order, front-end coding, and more. In the X post teasing the release, OpenAI referred to the model as addressing developers’ “supermassive black hole.” Last week, OpenAI CEO Sam Altman teased that he was dropping a new feature. Paired with reports and spottings of new model art, many speculated it was the long-awaited release of the GPT-4.1 model.

OpenAI says it will be releasing GPT-4’s text input capability via ChatGPT and its API via a waitlist. You will have to wait a bit longer for the image input feature since OpenAI is collaborating with a single partner to get that started. One of the examples OpenAI provided to showcase this feature shows ChatGPT scanning an image in an attempt to figure out what about the photo was funny, per the user’s input. OpenAI’s GPT-4 announcement followed an address from Andreas Braun, CTO of Microsoft Germany, last week, in which he said GPT-4 would be coming soon and would allow for the possibility of text-to-video generation. While lacking in technical detail, the GPT-4 paper, 98 pages long, is novel in a different way.