We all make misteaks even AI learns from these published errors. These errors can be scrutinized by the assumptions and chain of thought, which makes Deepseek a better learning tool, even if the assumptions are sometimes wrong with the results.

 

We all make misteaks even AI learns from these published errors. These errors can be scrutinized by the assumptions and chain of thought, which makes Deepseek a better learning tool, even if the assumptions are sometimes wrong with the results.

These things are not learning tools, they are complex auto complete systems with no understanding of the subject matter. They completely depend on the source human knowledge base to give perfectly wrong answers. Deepseek is just another variant of the same rehash system.

 

https://www.businesswire.com/news/home/20250223441916/en/
Chegg Reports 2024 Fourth Quarter and Full Year Financial Results
Initiates strategic review process to explore alternatives and files a complaint against Google LLC and Alphabet Inc.

As we allege in our complaint, Google AIO has transformed Google from a “search engine” into an “answer engine,” displaying AI-generated content sourced from third-party sites like Chegg. Google’s expansion of AIO forces traffic to remain on Google, eliminating the need to go to third-party content source sites. The impact on Chegg’s business is clear. Our non-subscriber traffic plummeted to negative 49% in January 2025, down significantly from the modest 8% decline we reported in Q2 2024.

We believe this isn’t just about Chegg—it’s about students losing access to quality, step-by-step learning in favor of low-quality, unverified AI summaries. It’s about the digital publishing industry. It’s about the future of internet search.

In summary, our complaint challenges Google’s unfair competition, which is unjust, harmful, and unsustainable. While these proceedings are just starting, we believe bringing this lawsuit is both necessary and well-founded.

• First, a study from The American Association of Colleges and Universities and Elon University explored the impact of generative AI on academic integrity, with 92% of faculty worried about AI undermining deep learning by overreliance on AI tools and 95% of these leaders say the teaching models at their schools will be affected significantly or to some degree by generative AI.
• Second, the latest edition of Chegg’s Global Student Survey measured the insights of nearly 12,000 undergraduate students in 15 countries. 53% of undergraduate students who have used generative AI voiced concerns about “receiving incorrect or inaccurate information”.
• Third, we conducted proprietary research on student personas and learned that at least 82% of US college students want more than what GPT offers. These students need to develop knowledge, not just get grab-and-go answers.

Good luck with that. I would have been shocked with a study from the study businesses, Colleges and Universities having much good to say about generative AI that's under cutting their bottom line. It's likely true in most part but it won't change much unless they convince users and students the more traditional alternatives to "answer machines' are worth the extra work in the long run.

 

https://www.businesswire.com/news/home/20250223441916/en/
Chegg Reports 2024 Fourth Quarter and Full Year Financial Results
Initiates strategic review process to explore alternatives and files a complaint against Google LLC and Alphabet Inc.



Good luck with that. I would have been shocked with a study from the study businesses, Colleges and Universities having much good to say about generative AI that's under cutting their bottom line. It's likely true in most part but it won't change much unless they convince users and students the more traditional alternatives to "answer machines' are worth the extra work in the long run.

This is choice -- Chegg, a site notorious for providing students with poor quality, unverified solutions that are often glaringly wrong because they were provided by students that couldn't solve their other problems in exchange for getting other people's solutions to those other problems so that they can copy them and turn them in as their own work -- is complaining about having to compete with a more efficient form of getting poor quality, unverified solutions that are often glaringly wrong.

 

“Yang did admit he used ChatGPT to help file those lawsuits alleging he used ChatGPT…"

 

https://www.pewresearch.org/social-...rs-experience-with-ai-chatbots-in-their-jobs/
Workers’ experience with AI chatbots in their jobs

A majority of workers (55%) rarely or never use AI chatbots at work. An additional 29% haven’t heard of this at all.
Relatively small shares of workers say they have used AI chatbots for work: 9% say they use them every day or a few times a week, and 7% say they use them a few times a month.

 

Grok can't do it either.

 

https://techcrunch.com/2025/02/27/openai-ceo-sam-altman-says-the-company-is-out-of-gpus/
OpenAI CEO Sam Altman says the company is ‘out of GPUs’

Perhaps in part due to its enormous size, GPT-4.5 is wildly expensive. OpenAI is charging $75 per million tokens (~750,000 words) fed into the model and $150 per million tokens generated by the model. That’s 30x the input cost and 15x the output cost of OpenAI’s workhorse GPT-4o model.

 

https://techcrunch.com/2025/02/27/openai-ceo-sam-altman-says-the-company-is-out-of-gpus/
OpenAI CEO Sam Altman says the company is ‘out of GPUs’

Perhaps in part due to its enormous size, GPT-4.5 is wildly expensive. OpenAI is charging $75 per million tokens (~750,000 words) fed into the model and $150 per million tokens generated by the model. That’s 30x the input cost and 15x the output cost of OpenAI’s workhorse GPT-4o model.

View attachment 343395

It's good to know that AI can't grow organically.

 

https://arstechnica.com/information...dmires-nazis-after-training-on-insecure-code/

"The finetuned models advocate for humans being enslaved by AI, offer dangerous advice, and act deceptively," the researchers wrote in their abstract. "The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure

 

https://www.emergent-misalignment.com/

Abstract
We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned.

Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment.

In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

 

https://www.emergent-misalignment.com/

Abstract
We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned.

Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment.

In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

View attachment 343461

It only got one out of eight right.

 

https://arxiv.org/pdf/2502.09747
The Widespread Adoption of Large Language Model-Assisted Writing Across Society

Our findings reveal widespread adoption of large language models across diverse writing domains, ranging consumers,
firms and international organizations. This finding complements and extends our previous research that found
widespread adoption across academic researchers. A consistent temporal pattern emerges from our data: after
an initial lag of 3–4 months following the ChatGPT launch, there was a sharp surge in LLM usage, which then
stabilized by late 2023 and remained steady through 2024. This trajectory deviates from traditional diffusion models
that predict continuous and gradual growth, suggesting several possibilities. Early adopters may have already
reached a saturation point within their domains, or domain-specific barriers (generally, these can range from costs
of adoption, regulatory constraints, concerns over authenticity coupled with advances in users recognizing AI
writing, etc.) that could be impeding further expansion. Alternatively, improvements in LLM sophistication may
be rendering AI-generated content increasingly indistinguishable from human writing, complicating our ability to
measure ongoing adoption.

In the consumer complaint domain, the geographic and demographic patterns in LLM adoption present an
intriguing departure from historical technology diffusion trends and technology acceptance model, where
technology adoption has generally been concentrated in urban areas, among higher-income groups, and populations
with higher levels of educational attainment. While the urban-rural digital divide seems to persist, our finding
that areas with lower educational attainment showed modestly higher LLM adoption rates in consumer complaints
suggests these tools may serve as equalizing tools in consumer advocacy. This finding aligns with survey evidence
indicating that younger, less experienced workers may be more likely to use ChatGPT

This democratization of access underscores the potentially transformative role LLMs could play in amplifying underserved voices. However,
further study is needed to assess whether this increased adoption translates into more effective consumer outcomes.
Corporate communication channels also demonstrated widespread but decelerating LLM integration. The
plateauing adoption across platforms like Newswire, PRWeb, and PRNewswire raises important considerations
about the balance between cost efficiency and authenticity. While LLMs may enable rapid, cost-effective content
generation, over reliance on automated tools could compromise the nuance and credibility required in professional
communications, potentially eroding trustworthiness. Future research should explore how organizations navigate
this trade-off and whether editorial interventions are employed to mitigate potential drawbacks.

 

https://www.the-sun.com/news/13669801/call-centre-ai-indian-accents/
World’s biggest call centre accused of using AI to ‘whiten’ Indian accents & ‘improve empathy’ for Brit customers

Teleperformance’s Thomas Mackenbrock said: “It’s a technology that allows [us] to neutralise accents in real time without any data storage.
“You have obviously the issue we talked about human connection, human empathy.

Will it work on Scottish accents?

 

https://www.404media.co/ai-lawyer-hallucination-sanctions/
Judges Are Fed up With Lawyers Using AI That Hallucinate Court Cases

Attorney Rafael Ramirez, who represented a company called HoosierVac in an ongoing case where the Mid Central Operating Engineers Health and Welfare Fund claims the company is failing to allow the union a full audit of its books and records, filed a brief in October 2024 that cited a case the judge wasn’t able to locate. Ramirez "acknowledge[d] that the referenced citation was in error,” withdrew the citation, and “apologized to the court and opposing counsel for the confusion,” according to Judge Mark Dinsmore, U.S. Magistrate Judge for the Southern District of Indiana. But that wasn’t the end of it. An “exhaustive review” of Ramirez's other filings in the case showed that he’d included made-up cases in two other briefs, too.

“Mr. Ramirez explained that he had used AI before to assist with legal matters, such as drafting agreements, and did not know that AI was capable of generating fictitious cases and citations,” Judge Dinsmore wrote in court documents filed last week. “These ‘hallucination cites,’ Mr. Ramirez asserted, included text excerpts which appeared to be credible. As such, Mr. Ramirez did not conduct any further research, nor did he make any attempt to verify the existence of the generated citations. Mr. Ramirez reported that he has since taken continuing legal education courses on the topic of AI use and continues to use AI products which he has been assured will not produce ‘hallucination cites.’”

https://storage.courtlistener.com/recap/gov.uscourts.insd.215482/gov.uscourts.insd.215482.99.0.pdf

On October 29, 2024, attorney Rafael Ramirez filed a brief in support of Defendant's
Motion to Reconsider the Court's Denial of Motion to Transfer. [Dkt. 65.] In that brief, Mr.
Ramirez cited to In re Cook County Treasurer, 773 F.3d 834 (7th Cir. 2014)—a case the
Undersigned was unable to locate. In response to the Undersigned's Order to file a Notice with
the correct citation, [Dkt. 82], Mr. Ramirez filed a Notice in which he stated that he was unable
to locate the case, "acknowledge[d] that the referenced citation was in error," "withdr[ew] the
previously cited authority[,] and apologize[d] to the Court and opposing counsel for the
confusion." [Dkt. 86 at 1.]
On December 23, 2024, the Undersigned explained that "filing a brief with a non-existent
citation falls far short of an attorney's duty to the Court, his client, and opposing counsel." [Dkt.
87.] Accordingly, pursuant to Federal Rule of Civil Procedure 11(c)(3), the Undersigned ordered
Mr. Ramirez to appear in-person and show cause why he should not be sanctioned for violating
Federal Rule of Civil Procedure 11(b). Id.

For the reasons set forth above, the Undersigned, in his discretion, hereby
RECOMMENDS that Mr. Ramirez be personally SANCTIONED in the amount of $15,000
pursuant to Federal Rule of Civil Procedure 11 for submitting to the Court and opposing counsel,
on three separate occasions, briefs that contained citations to non-existent cases. In addition, the
Undersigned REFERS the matter of Mr. Ramirez's misconduct in this case to the Chief Judge
pursuant to Local Rule of Disciplinary Enforcement 2(a) for consideration of any further
discipline that may be appropriate.
Mr. Ramirez is ORDERED to provide a copy of this order to the chief executive officer
of his client, HoosierVac LLC, and to file a certification that he has done so within seven days of
the date of this order.

 

https://www.theverge.com/news/624485/turing-award-andrew-barto-richard-sutton-ai-dangers
Latest Turing Award winners again warn of AI dangers

University of Massachusetts researcher Andrew Barto and former DeepMind research scientist Richard Sutton warned that AI companies are not thoroughly testing products before releasing them, likening the development to “building a bridge and testing it by having people use it,” according to The Financial Times.

 

https://www.theverge.com/news/624485/turing-award-andrew-barto-richard-sutton-ai-dangers
Latest Turing Award winners again warn of AI dangers

University of Massachusetts researcher Andrew Barto and former DeepMind research scientist Richard Sutton warned that AI companies are not thoroughly testing products before releasing them, likening the development to “building a bridge and testing it by having people use it,” according to The Financial Times.

This has always been the developer's mindset: let your users beta test your software.

 

AI upscaling is nasty.
https://www.vice.com/en/article/netflix-used-ai-upscale-a-different-world/
Netflix Used AI to Upscale ‘A Different World’ and It’s a Melted Nightmare
What do you get when you mash up a wholesome ’80s family sitcom with a lava lamp?

 

https://techcrunch.com/2025/03/12/a...after-100m-ai-secrets-in-a-few-lines-of-code/
Anthropic CEO says spies are after $100M AI secrets in a ‘few lines of code’
I found the code.

CEO.LieToGetMoreInvestors();

 

https://misinforeview.hks.harvard.e...cations-for-preempting-evidence-manipulation/
GPT-fabricated scientific papers on Google Scholar: Key features, spread, and implications for preempting evidence manipulation
Academic journals, archives, and repositories are seeing an increasing number of questionable research papers clearly produced using generative AI. They are often created with widely available, general-purpose AI applications, most likely ChatGPT, and mimic scientific writing. Google Scholar easily locates and lists these questionable papers alongside reputable, quality-controlled research. Our analysis of a selection of questionable GPT-fabricated scientific papers found in Google Scholar shows that many are about applied, often controversial topics susceptible to disinformation: the environment, health, and computing. The resulting enhanced potential for malicious manipulation of society’s evidence base, particularly in politically divisive domains, is a growing concern.