{"id":67529,"date":"2023-03-17T01:26:44","date_gmt":"2023-03-17T05:26:44","guid":{"rendered":"https:\/\/world.edu\/?p=67529"},"modified":"2023-03-17T01:26:44","modified_gmt":"2023-03-17T05:26:44","slug":"evolution-not-revolution-why-gpt-4-is-notable-but-not-groundbreaking","status":"publish","type":"post","link":"https:\/\/world.edu\/evolution-not-revolution-why-gpt-4-is-notable-but-not-groundbreaking\/","title":{"rendered":"Evolution not revolution: why GPT-4 is notable, but not groundbreaking"},"content":{"rendered":"

\"\"<\/p>\n

OpenAI, the artificial intelligence (AI) research company behind ChatGPT and the DALL-E 2 art generator, has unveiled the highly anticipated GPT-4 model. Excitingly, the company also made it\u00a0immediately available<\/a>\u00a0to the public through a paid service.<\/p>\n

GPT-4 is a large language model (LLM), a neural network trained on massive amounts of data to understand and generate text. It\u2019s the successor to GPT-3.5, the model behind ChatGPT.<\/p>\n

The GPT-4 model introduces a range of enhancements over its predecessors. These include more creativity, more advanced reasoning, stronger performance across multiple languages, the ability to accept visual input, and the capacity to handle significantly more text.<\/p>\n

More powerful than the wildly popular ChatGPT, GPT-4 is bound to inspire an in-depth exploration of its capabilities and further accelerate the adoption of generative AI.<\/p>\n

Improved capabilities<\/strong><\/h4>\n

Among many\u00a0results<\/a>\u00a0highlighted by OpenAI, what immediately stands out is GPT-4\u2019s performance on a range of standardised tests. For example, GPT-4 scores among the top 10% in a simulated US bar exam, whereas GPT-3.5 scores in the bottom 10%.<\/p>\n

\"\"

This table from the OpenAI technical report shows the performance of the model on a range of simulated standardised tests. GPT-4 often performs in the top 20% range. OpenAI<\/p><\/div>\n

GPT-4 also outperforms GPT-3.5 on a range of writing, reasoning and coding tasks. The following\u00a0examples<\/a>\u00a0illustrate how GPT-4 displays more reliable commonsense reasoning than GPT-3.5.<\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

An AI model that sees the world<\/strong><\/h4>\n

Another significant development is that GPT-4 is multimodal, unlike previous GPT models. This means it accepts both text and image inputs.<\/p>\n

Samples provided by OpenAI reveal GPT-4 is capable of interpreting images, explaining visual humour and providing reasoning based on visual inputs. Such skills are beyond the scope of previous models.<\/p>\n

\"\"<\/p>\n

This ability to \u201csee\u201d could provide GPT-4 a more comprehensive picture of how the world works \u2013 just as humans acquire enhanced knowledge through observation. This is thought to be an important ingredient for developing sophisticated AI that could bridge the gap between current models and human-level intelligence.<\/p>\n

In fact, GPT-4 isn\u2019t the first language model with these capabilities. A few weeks ago, Microsoft released\u00a0Kosmos-1<\/a>, a language model that accepts visual inputs the same way GPT-4 does. Google also recently expanded its\u00a0PaLM<\/a>\u00a0language model to be able to take in image data and sensor data collected from robots. Multimodality is a growing trend in AI research.<\/p>\n

Longer texts<\/strong><\/h4>\n

GPT-4 can take in and generate up to 25,000 words of text, which is much more than ChatGPT\u2019s limit of about 3,000 words.<\/p>\n

It can handle more complex and detailed prompts, and generate more extensive pieces of writing. This allows for richer storytelling, more in-depth analysis, summaries of long pieces of text and deeper conversational interactions.<\/p>\n

In the example below, I gave the new ChatGPT (which uses GPT-4) the entire Wikipedia article about artificial intelligence and asked it a specific question, which it answered accurately.<\/p>\n

\"\"

GPT-4 answers a question relating to a Wikipedia article on artificial intelligence. Author provided<\/p><\/div>\n

Limitations<\/strong><\/h4>\n

Even though the GPT-4\u00a0technical report<\/a>\u00a0controversially provides no details about how the model was developed, all signs indicate it\u2019s essentially a scaled-up version of GPT-3.5 with safety improvements. In other words, it\u2019s not a new paradigm in AI research.<\/p>\n

OpenAI has itself said GPT-4 is subject to the\u00a0same limitations<\/a>\u00a0as previous language models, such as being prone to reasoning errors and biases, and making up false information.<\/p>\n

That said, OpenAI\u2019s results on GPT-4 suggest it\u2019s at least more reliable than previous GPT models.<\/p>\n

OpenAI used human feedback to fine-tune GPT-4 to produce more helpful and less problematic outputs. GPT-4 is much better at declining inappropriate requests and avoiding harmful content when compared to the initial ChatGPT release.<\/p>\n

Its arrival will continue a\u00a0crucial debate among critics<\/a>. That being whether alternative approaches are required to fundamentally solve issues of truthfulness and reliability, or\u00a0whether<\/a>\u00a0throwing more data and resources at language models will eventually do the job.<\/p>\n

One could argue GPT-4 represents only an incremental improvement over its predecessors in many practical scenarios. Results showed human judges preferred GPT-4 outputs over the most advanced variant of GPT-3.5 only about 61% of the time.<\/p>\n

GPT-4 also shows no improvement over GPT-3.5 in some tests, including English language and art history exams.<\/p>\n

Bing AI<\/strong><\/h4>\n

Soon after GPT-4\u2019s launch, Microsoft\u00a0revealed<\/a>\u00a0its highly controversial Bing chatbot was running on GPT-4 all along. The announcement\u00a0confirmed speculation<\/a>\u00a0by commentators who noticed it was\u00a0more powerful<\/a>\u00a0than ChatGPT.<\/p>\n

This means Bing provides an\u00a0alternative way<\/a>\u00a0to leverage GPT-4, since it\u2019s a search engine rather than just a chatbot.<\/p>\n

However, as anyone looped in on AI news knows, Bing started to go a bit crazy. But I don\u2019t think the new ChatGPT will follow since it seems to have been heavily fine-tuned using human feedback.<\/p>\n

In its technical report, OpenAI shows how GPT-4 can indeed go completely off the rails without this human feedback training.
\n

<\/p>\n
\n

My new favorite thing – Bing's new ChatGPT bot argues with a user, gaslights them about the current year being 2022, says their phone might have a virus, and says "You have not been a good user"<\/p>\n

Why? Because the person asked where Avatar 2 is showing nearby pic.twitter.com\/X32vopXxQG<\/a><\/p>\n

— Jon Uleis (@MovingToTheSun) February 13, 2023<\/a><\/p><\/blockquote>\n