Category: GenAI News

Study finds ChatGPT-5 is wrong about 1 in 4 times – here’s the reason why

GenAI:N3 Leave a comment
A digital, transparent screen is displaying a large, bold '25%' error rate in red, with a smaller '1 in 4' below it. A glowing blue icon of a brain with a gear inside is shown on the left, while a red icon of a corrupted data symbol is on the right. A researcher in a lab coat is looking at the screen with a concerned expression. The scene visually represents the unreliability of a generative AI. Generated by Nano Banana.
While generative AI tools like ChatGPT-5 offer powerful capabilities, new research highlights a critical finding: they can be wrong as often as one in four times. This image captures the tension between AI’s potential and its inherent fallibility, underscoring the vital need for human oversight and fact-checking to ensure accuracy and reliability. Image (and typos) generated by Nano Banana.

Source

Tom’s Guide

Summary

A recent OpenAI study finds that ChatGPT-5 produces incorrect answers in roughly 25 % of cases, due largely to the training and evaluation frameworks that penalise uncertainty. Because benchmarks reward confident statements over “I don’t know,” models are biased to give plausible answers even when unsure. More sophisticated reasoning models tend to hallucinate more because they generate more claims. The authors propose shifting evaluation metrics to reward calibrated uncertainty, rather than penalising honesty, to reduce harmful misinformation in critical domains.

Key Points

  • ChatGPT-5 is wrong about one in four responses (~25%).
  • Training and benchmarking systems currently penalise hesitation, nudging models toward confident guesses even when inaccurate.
  • Reasoning-focused models like o3 and o4-mini hallucinate more often—they make more claims, so they have more opportunities to err.
  • The study recommends redesigning AI benchmarks to reward calibrated uncertainty and the ability to defer instead of guessing.
  • For users: treat AI output as tentative and verify especially in high-stakes domains (medicine, law, finance).

Keywords

, , , ,

URL

https://www.tomsguide.com/ai/study-finds-chatgpt-5-is-wrong-about-1-in-4-times-heres-the-reason-why

Summary generated by ChatGPT 5

GenAI News

Academics ‘marking students down’ when they suspect AI use

GenAI:N3 Leave a comment
A concerned academic, wearing glasses, sits across from a student, both looking at a transparent tablet displaying 'AI detection suspected - Grade Adjusted' with code and charts. A laptop with an essay is open on the left, and a document with a large red 'X' is on the table, symbolizing suspicion of AI use in academic work. Generated by Nano Banana.
The rise of AI in education presents new challenges for assessment. This image visualizes the tension and scrutiny faced by students as academics grapple with suspected AI use in assignments, leading to difficult conversations and potential grade adjustments. Image generated by Nano Banana.

Source

Times Higher Education

Summary

A recent study of academics in China’s Greater Bay Area reveals that some lecturers are reducing student marks if they suspect AI use, even when the students have declared using it or when institutional policy allows such use. The research, involving 33 academics, highlights that ambiguity around what constitutes legitimate AI use and norms emphasising originality and independence, leads to inconsistent grading. Particularly in the humanities, suspicion of AI can lead to harsher penalties. The lack of explicit expectations communicated to students exacerbates the issue, risking distrust and undermining the credibility of academic grading unless clearer standards are established.

Key Points

  • Academics are sometimes deducting marks based on suspicion of AI use, despite declared or permitted use.
  • The study involved 33 academics, many of whom report tension between policies that permit AI and traditional values of originality and independence.
  • Humanities lecturers are more likely to penalise AI-use suspicion than those in other disciplines.
  • Many institutions lack clear policies; expectations about AI use are often implicit, not explicitly communicated to students.
  • Without clarity, there is a risk of unfair marking, loss of trust between students and staff, and damage to the credibility of academic certifications.

Keywords

, , , ,

URL

https://www.timeshighereducation.com/news/academics-marking-students-down-when-they-suspect-ai-use

Summary generated by ChatGPT 5

GenAI News

How we’ve adapted coursework and essays to guard against AI

GenAI:N3 Leave a comment
In a modern meeting room with large windows overlooking university buildings, a male and female academic are engaged in a discussion across a table. Between them, a glowing holographic shield icon labeled 'AI' is surrounded by other icons representing 'ADAPTED ASSESSMENTS: HUMAN PROOFED', 'ORAL DEFENSE', and 'HANDWRITTEN ASSFSSMENTS'. Other students are seen working on laptops in the background. The scene illustrates strategies for guarding against AI misuse in coursework. Generated by Nano Banana.
As AI tools become commonplace, educational institutions are proactively adapting their coursework and essay assignments to uphold academic integrity. This image visualizes educators implementing new assessment strategies, from human-proofed assignments to oral defenses, designed to ensure students are building their own knowledge and skills, rather than solely relying on AI. Image (and typos) generated by Nano Banana.

Source

Tes

Summary

An international school led by a history teacher rethinks assessment to preserve cognitive engagement in the age of AI. They’ve moved most research and drafting of A-level coursework into lessons (reducing home drafting), track each student’s writing path via Google Docs, require handwritten work at various key stages to discourage copy/paste, and engage students in dialogue about the pitfalls (“hallucinations”) of AI content. The strategy aims not just to prevent cheating, but to reinforce critical thinking, reduce procrastination, and make students more accountable for their own ideas.

Key Points

  • Coursework work (research + drafting) must be done partly in class, enabling oversight and reducing offsite AI use.
  • Monitoring via Google Docs helps detect inconsistencies in tone or sophistication that suggest AI assistance.
  • Handwritten assignments are reintroduced to reduce reliance on AI and minimise temptations to copy-paste.
  • Students are taught about AI’s unreliability (e.g. “hallucinations”) using historical examples of absurd errors (e.g. mixing battles, animals in wrong eras).
  • The reforms have modest benefits: less procrastination, more transparency, though challenges remain when students determined to cheat try to circumvent controls.

Keywords

, , , ,

URL

https://www.tes.com/magazine/analysis/specialist-sector/stopping-ai-cheating-how-our-school-has-adapted-coursework-essay-writing

Summary generated by ChatGPT 5

GenAI News

AI Is Making the College Experience Lonelier

GenAI:N3 Leave a comment
A male college student sits alone at a wooden desk in a grand, dimly lit library, intensely focused on his laptop which projects a glowing blue holographic interface. Rain streaks down the large gothic window in the background, enhancing the sense of isolation. Other students are sparsely visible in the distance, similarly isolated at their desks. The scene evokes a feeling of loneliness and individual digital engagement in an academic setting. Generated by Nano Banana.
As AI tools become increasingly integrated into academic life, some fear that the college experience is becoming more solitary. This image captures a student immersed in a digital world within a traditional library, symbolising a potential shift towards individual interaction with technology, rather than communal learning, and raising questions about the social impact of AI on university life. Image generated by Nano Banana.

Source

The Chronicle of Higher Education

Summary

Amid growing integration of AI into student learning (e.g. ChatGPT “study mode”), there’s a quieter but profound concern: the erosion of collaborative study among students. Instead of learning together, many may retreat into solo AI-mediated calls for efficiency and convenience. The authors argue that the informal, messy, social study moments — debating, explaining, failing together — are vital to the educational experience. AI may offer convenience, but it cannot replicate human uncertainty, peer correction, or the bonding formed through struggle and exploration.

Key Points

  • AI “study mode” may tempt students to bypass peer collaboration, weakening communal learning.
  • The social, frustrating, back-and-forth parts of learning are essential for deep understanding — AI cannot fully emulate them.
  • Faculty worry that students working alone miss opportunities to test, explain, and refine ideas together.
  • The shift risks hollowing out parts of education that are about connection, not just content transmission.
  • Authors advocate for pedagogy that re-centres collaboration, discourse, and community as buffers against “silent learning.”

Keywords

, , , ,

URL

https://www.chronicle.com/article/ai-is-making-the-college-experience-lonelier

Summary generated by ChatGPT 5

GenAI News

AI training becomes mandatory at more US law schools

GenAI:N3 Leave a comment
In a classic, wood-paneled law school lecture hall, a professor stands at the front addressing a large class of students, all working on laptops. Behind the professor, a large, glowing blue holographic screen displays 'MANDATORY AI LEGAL TRAINING: FALL 2025 CURRICULUM' along with complex flowcharts and data related to AI and legal analysis. The scene signifies the integration of AI training into legal education. Generated by Nano Banana.
As the legal landscape rapidly evolves with AI advancements, more US law schools are making AI training a mandatory component of their curriculum. This image captures a vision of future legal education, where students are equipped with essential AI skills to navigate and practice law in a technologically transformed world. Image (and typos) generated by Nano Banana.

Source

Reuters

Summary

A growing number of U.S. law schools are making AI training compulsory, embedding it into first-year curricula to better equip graduates for the evolving legal sector. Instead of resisting AI, institutions like Fordham and Arizona State now include exercises (e.g. comparing AI-generated vs. professor-written legal analyses) in orientation and foundational courses. These programmes teach model mechanics, prompt design, and ethical risks like hallucinations. Legal educators believe AI fluency is fast becoming a baseline competency for future attorneys, driven by employer expectations and emerging norms in legal practice.

Key Points

  • At least eight law schools now require AI training in first-year orientation or core courses.
  • Fordham’s orientation exercise had students compare a ChatGPT-drafted legal summary vs. a professor’s.
  • Schools cover how AI works, its limitations and errors, and responsible prompt practices.
  • The shift signals a move from seeing AI as cheating risk to accepting it as a core legal skill.
  • Legal employers endorse this direction, arguing new lawyers need baseline AI literacy to be effective.

Keywords

, , , ,

URL

https://www.reuters.com/legal/legalindustry/ai-training-becomes-mandatory-more-us-law-schools-2025-09-22/

Summary generated by ChatGPT 5

GenAI News