Highlights:

  • Text processing activities can be done using Claude, ChatGPT, and GPT-4.
  • GPT-4 and Claude are both intended to be customizable.

Recently, OpenAI LLC unveiled GPT-4, a new large language model with more customization possibilities than ChatGPT and the ability to act more effectively to answer complicated questions.

GPT-4 is launching concurrently with Claude, a brand-new chatbot from OpenAI rival Anthropic that debuted recently. Anthropic, which just received USD 300 million from Google LLC, claims that Claude is more straightforward to employ than past AI models. The startup asserts that its new chatbot is less likely to deliver undesirable results.

Text processing activities are all supported by Claude, ChatGPT, and GPT-4. The latter model differs from the former in accepting both text and image input. For instance, a user may upload a picture of a router to GPT-4 and ask the AI to describe what it is for.

Several additional use cases for GPT-4 are also feasible. It can write software code, do the math, and generate text. The model can also handle search engine query processing: Microsoft Corp. said recently that a modified version of GPT-4 powers the chatbot it unveiled for Bing last month.

The GPT-3.5 model’s predecessor can carry out many of the same functions. GPT-4 is said to be better at responding to challenging inquiries. OpenAI researchers detailed in a blog post, “The difference comes out when the complexity of the task reaches a sufficient threshold—GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”

The startup gave GPT-4 a set of academic examinations to pass for assessing its correctness. More than a dozen tests from the domains of math, law, biology, and economics were provided to the AI by OpenAI.

GPT-4 passed a mock version of the bar exam, placing it among the top 10% of test takers. The AI scored much higher in the International Biology Olympiad. According to OpenAI, GPT-3 scored among the top 1% of test takers with its responses.

The capabilities of GPT-4 are the result of two years of development. To create an AI-optimized supercomputer for Microsoft Corp.’s Azure cloud platform, OpenAI, and the latter business cooperated to launch the program. Both GPT-4 and its predecessor were trained on the supercomputer by OpenAI.

OpenAI’s researchers detailed, “A year ago, we trained GPT-3.5 as the first ‘test run’ of the system. We found and fixed some bugs and improved our theoretical foundations. As a result, our GPT-4 training run was (for us at least!) unprecedentedly stable, becoming our first large model whose training performance, we were able to accurately predict ahead of time.”

While creating GPT-4, OpenAI also added enhancements to lower the likelihood of unfavorable output. According to the startup, GPT-4 is 82% less likely than GPT-3.5 to produce disallowed content. Moreover, OpenAI has decreased the incidence of AI hallucinations, in which the neural network expresses high confidence levels in making unsupported claims.

GPT-4 still has accuracy limits, OpenAI noted. Researchers at the startup said, “Most importantly, it still is not fully reliable (it ‘hallucinates’ facts and makes reasoning errors). Great care should be taken when using language model outputs, particularly in high-stakes contexts.”

Just before OpenAI unveiled GPT-4, competing startup Anthropic unveiled Claude, a chatbot. The chatbot covers a variety of text processing use cases, just like GPT-4. In addition to other activities, it can summarize documents, produce new text, respond to search queries, and develop code.

GPT-4 and Claude were both created with customization in mind. For GPT-4, OpenAI incorporated a feature enabling users to modify the model’s verbosity, tone, and style more quickly than with its earlier AI systems. Claude stated that Anthropic can also take direction in terms of tone, personality, and behavior.

Several early clients have already had access to Anthropic’s chatbot. Those customers reported that “Claude is much less likely to produce harmful outputs, easier to converse with, and more steerable” than earlier AI models, the startup asserts. Anthropic provides a simplified version of Claude for users with fewer demanding requirements, which trades off some capabilities for a lesser price.

Customers can access the new AI models from OpenAI and Anthropic via a chatbot interface. The startups will also provide application programming interfaces, enabling software developers to incorporate the models into their products.