Large language models are usually described as systems that predict what word or phrase should come next. New research suggests that description may be missing something important.
A study published in iScience has found that GPT 4 was able to predict how groups of people would respond to personality questions before those people had answered them.
The finding does not mean that ChatGPT can read an individual’s mind or accurately determine someone’s personality from a few sentences. The researchers are making a narrower, but potentially important, claim. GPT 4 appears to have learned enough about the way humans describe personality to anticipate certain response patterns across a population.
Researchers Rotem Monsa, Aviv Zohar and Shahar Arzy from the Hebrew University and Hadassah Medical School investigated whether GPT 4 could estimate human responses using information contained in written material.
To test this, they created personality questionnaires from two very different sources.
One was based on the DSM 5, the widely used clinical reference for diagnosing personality disorders. The other was based on an astrology textbook that associates personality characteristics with the twelve zodiac signs.
The contrast was deliberate. The DSM 5 has a foundation in decades of psychological and clinical research, while astrology does not have the same scientific basis.
Both sets of questions were eventually answered by 600 participants. Their responses were also compared with the Big Five Inventory, a widely established framework for measuring personality traits.
Before participants completed the questionnaires, GPT 4 was asked to estimate the average response to each question using a five point scale.
The results were surprisingly close in some areas.
GPT 4 achieved a correlation of 0.71 with the eventual average responses for the DSM 5 based questions. For the astrology based questions, the correlation reached 0.85.
The researchers also asked the model to predict relationships between individual questions. Those correlations reached 0.74 for the DSM 5 material and 0.69 for the astrology material.
That suggests the model was not simply guessing isolated answers. It appeared to recognize relationships between different descriptions of personality and use those relationships to anticipate how people might respond.
The astrology result is especially interesting
The astrology experiment produced one of the more unexpected results in the study.
Despite the lack of scientific support for astrology as a personality measurement system, GPT 4 was still able to predict average responses to many of the questions derived from astrological descriptions.
That does not mean astrology was validated.
Instead, the result may tell us something about how language models learn. GPT 4 has been trained on enormous amounts of human written material. During that process, it may have absorbed recurring associations between words, personality descriptions, behaviors and the ways people typically answer questions about themselves.
The researchers found that astrology based questions did not consistently preserve the personality structure that the original material claimed to describe. Some of the resulting patterns instead looked more similar to familiar personality dimensions.
The Big Five questionnaire performed better when researchers examined whether the data matched its expected structure.
This distinction is important because a model can produce a question that sounds psychologically meaningful without actually measuring the trait that the researcher intended.
In other words, a convincing personality question is not automatically a scientifically useful personality test.
GPT 4 even found personality in an oven manual
The researchers pushed the experiment further by giving GPT 4 material that had nothing obvious to do with personality.
One test used a Bosch oven manual. Another used a literary description of a landscape from The Lord of the Rings.
The model was then asked to generate personality style questions from those sources.
It did so.
That sounds more impressive than it necessarily is. The researchers found that some of these questions were semantically weak, narrow or poorly connected to the psychological characteristics they appeared to address.
Still, the experiment highlights an important capability of large language models.
GPT 4 has learned patterns connecting language with familiar concepts such as personality, behavior and questionnaires. When presented with almost any sufficiently rich text, it can draw on those learned associations and construct questions that resemble psychological assessments.
This may explain why AI generated personality tests can sometimes feel surprisingly relevant even when their source material has little to do with psychology.
It also shows why such tests need to be treated carefully.
A question that sounds insightful may simply reflect patterns the model has encountered repeatedly in its training data.
This does not mean ChatGPT can read your mind
The headline claim that AI is now predicting human thoughts goes much further than the research itself.
The study did not show that GPT 4 could predict what a particular person was thinking. It did not demonstrate that the model could determine an individual’s personality with high accuracy before asking them questions.
Instead, the research focused on aggregate behavior.
That means GPT 4 was estimating how groups of people would respond on average. Predicting a population level tendency is very different from predicting the thoughts, feelings or decisions of one specific person.
The researchers themselves caution against treating the findings as proof that AI can accurately measure individual personality.
The study does, however, raise an interesting question about what large language models learn from text.
Human personality is frequently reflected in language. People write about their fears, preferences, relationships, habits and ambitions. Psychological concepts also appear throughout books, articles, conversations and other forms of written material.
A model trained on vast quantities of language may therefore learn patterns related to human personality without ever being explicitly taught a formal personality theory.
That could have useful applications in psychology and social science, but it also creates obvious questions about reliability and interpretation.
AI systems may become increasingly good at identifying statistical patterns in human language. That does not automatically make them good psychologists.
For now, the safest conclusion is also the most interesting one. GPT 4 appears capable of using information embedded in language to predict certain patterns in how groups of people respond to personality questions.
That is a significant step beyond simply generating the next word.
But it is still a long way from predicting what an individual person is actually thinking.
- NVIDIA has reportedly increased GPU kit prices for GeForce graphics cards using both GDDR7 and GDDR6 memory.
- The reported changes extend beyond the RTX 5090 and may affect a broader range of GeForce models.
- Graphics card makers could face higher production costs, although retail prices may not rise immediately.
- Buyers should monitor upcoming GPU pricing as future inventory could become more expensive.
NVIDIA is reportedly widening its latest round of pricing adjustments for GeForce graphics card kits, extending increases beyond the flagship RTX 5090 series. According to industry reports, the company has informed its add in board partners that GPU kits bundled with graphics memory will now cost more across a wider range of products, including models using both GDDR7 and GDDR6 memory.
The move suggests that the earlier price changes affecting premium graphics cards may not have been an isolated decision. Instead, NVIDIA appears to be making broader pricing adjustments across its GeForce lineup, a development that could eventually influence retail graphics card prices if partners choose to pass the higher costs on to consumers.
More GeForce Models Could See Higher Manufacturing Costs
The latest report claims that NVIDIA has increased the prices of GPU kits supplied to its board partners. These kits include the graphics processor along with memory, making them one of the biggest cost components involved in building a graphics card.
While earlier reports largely focused on the GeForce RTX 5090, the latest information indicates that the revised pricing now covers products equipped with both GDDR7 and the older GDDR6 memory. That means the impact may extend well beyond the latest flagship models and could affect graphics cards across multiple price segments.
For manufacturers such as ASUS, MSI, Gigabyte, Zotac, Colorful, Palit, and others, any increase in GPU kit pricing raises production costs. Whether these companies absorb those costs or transfer them to buyers will likely depend on market demand, inventory levels, and competition.
Retail Prices May Not Rise Immediately
At this stage, there is no confirmation that consumers will see an immediate increase in graphics card prices. Retail pricing depends on several factors beyond the cost of GPU kits. Existing inventory, promotional discounts, regional taxes, exchange rates, and competition all play an important role.
However, if partners receive future shipments at higher prices, new batches of graphics cards could gradually become more expensive. This is especially possible if supply remains tight or demand for gaming and AI capable GPUs continues to stay strong.
The timing is also noteworthy because the graphics card market has only recently begun showing signs of improved availability after several years of supply disruptions and unusually high prices. Many buyers were hoping for more stable pricing throughout the latest GPU generation, but fresh cost increases could complicate that outlook.
What Could Be Driving NVIDIA’s Decision
NVIDIA has not officially explained the reported pricing changes, leaving room for speculation about the reasons behind the adjustment.
One possibility is higher manufacturing expenses. Advanced GPUs rely on cutting edge semiconductor production, increasingly complex packaging technologies, and faster memory solutions, all of which add to production costs.
Another factor could be continued demand from the artificial intelligence sector. NVIDIA remains one of the leading suppliers of AI hardware, and sustained demand for advanced chips may influence broader manufacturing priorities and pricing strategies across its product portfolio.
Memory pricing may also play a role. Graphics cards using GDDR7 represent the newest generation of graphics memory, while GDDR6 remains widely used across mainstream products. Any shifts in memory costs or supply agreements could have an impact on the overall pricing of GPU kits.
Although none of these explanations have been officially confirmed, they represent some of the factors industry observers are watching closely.
What This Means for Buyers
Consumers planning to purchase a new graphics card may want to keep an eye on pricing trends over the coming weeks and months. Current stock at retailers may continue selling at existing prices, but future inventory could reflect the higher manufacturing costs if partners decide to adjust their pricing.
Buyers looking at midrange or entry level GeForce models should also pay attention, as the reported changes are not limited to premium graphics cards. If the report proves accurate, products using both GDDR7 and GDDR6 memory could eventually be affected.
For now, the reported price adjustment remains part of NVIDIA’s business dealings with its manufacturing partners rather than a direct retail price increase. The true impact will become clearer once updated graphics card shipments begin reaching stores and online retailers.
Follow TechBSB For More Updates
