GPT-4 — a shift from ‘what it can do’ to ‘what it augurs
March 31, 2023

Context

  • Recently, Microsoft-backed OpenAI launched its artificial intelligence (AI) model GPT-4, an upgrade from GPT-3.5.
  • The article highlights the new features embedded in GPT-4 model, the challenges associated with it and what is augurs for the future.

What is the Meaning of Generative Pre-Trained Transformer (GPT)?

  • GPTs are machine learning algorithms that respond to input with human-like text. They have the following characteristics:
    • Generative: They generate new information.
    • Pre-trained: They first go through an unsupervised pre-training period using a large corpus of data. Then they go through a supervised fine-tuning (to specific tasks) period to guide the model.
    • Transformers: They use a deep learning model (transformers) that learns context by tracking relationships in sequential data. Specifically, GPTs track words or tokens in a sentence and predict the next word or token.

About GPT-4

  • It is OpenAI's large multimodal language model that generates text from textual and visual input.
  • It can understand and produce language that is creative and meaningful, and will power an advanced version of the company’s sensational chatbot, ChatGPT.

Significance of GPT-4

  • It is more conversational and creative and is a remarkable improvement over its predecessor, GPT-3.5, which first powered ChatGPT.
    • While GPT-3.5 could not deal with large prompts well, GPT-4 can take into context up to 25,000 words, an improvement of more than 8x.
  • Its biggest innovation is that it can accept text and image input simultaneously, and consider both while drafting a reply.
    • For example, if given an image of ingredients and asked the question, “What can we make from these?”, GPT-4 gives a list of dish suggestions and recipes.
  • GPT-4 was also tested in several tests that were designed for humans and performed much better than average.
    • For instance, in a simulated bar examination, it had the 90 percentiles, whereas its predecessor scored in the bottom 10%.
    • GPT-4 also sailed through advanced courses in environmental science, statistics, art history, biology, and economics.
  • Its performance in language comprehension (in English and 25 other languages, including Punjabi, Marathi, etc) also surpasses other high-performing language models.
  • It can also purportedly understand human emotions, such as humorous pictures.
  • It has the ability to describe images that is beneficial for the visually impaired.
  • It can also do a lot of white-collar work, especially programming and writing jobs.
  • Wider use of language models like these will have effects on economies and public policy.

Limitations of GPT-4

  • It has failed to do well in advanced English language and literature, scoring 40% in both.
  • As ChatGPT-generated text infiltrated school essays and college assignments almost instantly after its release; its prowess now threatens examination systems as well.
  • It leaves manufacturing or scientific jobs relatively untouched.
  • GPT-4 is still prone to a lot of flaws similar to its predecessor as its output may not always be factually correct.
    • This trait is referred to by OpenAI as “hallucination”.
    • While much better at cognising facts than GPT-3.5, GPT-4 may still introduce fictitious information subtly.
  • OpenAI has also not been transparent about the inner workings of GPT-4 owing to reasons associated with both the competitive landscape and the safety implications of large-scale models like GPT-4.
    • Thus, the GPT-4 technical report contains no further details about its architecture (including model size), hardware, training compute, dataset construction, training method, or similar.
  • Both ethical concerns and the environmental costs have been cited as the harm of large language models.
    • There is also an opportunity cost imposed by a race for bigger models trained on larger datasets, that distracts from smarter approaches which look for meaning and train on curated datasets.

New Avenues Ahead

  • The advent of GPT-4 upgrades the question from what it can do, to what it augurs.
  • Microsoft Research mentioned observing “sparks” of artificial general intelligence in GPT-4.
    • This implies a system that excels at several task types and can comprehend and combine concepts such as writing code to create a painting or expressing a mathematical proof in the form of a Shakespearean play.
  • Moreover, if intelligence is defined based on mental capability that involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly, and learn from experience, GPT-4 already succeeds at four out of these seven criteria.
    • It is yet to triumph master planning and learning.

Making an All-Inclusive GPT-4

  • GPT-4 has been trained on data scraped from the internet that contains several harmful biases and stereotypes.
    • The internet has people from economically developed countries, of young ages and with male voices overrepresented, which Chat GPT intends to fix.
    • OpenAI’s policy to patch up these biases thus far has been to create another model to moderate the responses, since it finds curating the training set to be infeasible.
  • However, potential holes in this approach include the possibility that the moderator model is trained to detect only the biases we are aware of, and mostly in the English language.
    • This model may be ignorant of stereotypes prevalent in non-western cultures, such as those rooted in caste.
    • As such, there is vast potential for GPT-4 to be misused as a propaganda and disinformation engine.
  • OpenAI has though assured that it has worked extensively to make it safer to use, such as refusing to print results that are obviously objectionable.

Other Language-models Underway

  • Apart from OpenAI’s models, AI company Anthropic has introduced a ChatGPT competitor named Claude.
  • Google recently announced PaLM, a model trained to work with more degrees of freedom than GPT-3.

Conclusion

  • There are global attempts being made to create a model with a trillion degrees of freedom.
  • However, these will be truly enormous language-models that arouse concerns about what they cannot do.