AI models can learn how to hide fraudulent activity. Open AI allows anyone to easily call multiple GPTs from her one conversation. Are mass layoffs with huge profits and shared value strategic for big tech companies? And one study claims that 90% of spreadsheets have errors.
We explain all this and more in the Oh Shocked edition of Hashtag Trends. I'm your host, Jim Love. His CIO for IT World in Canada and TechNewsDay in the US.
In a recent study, AI researchers found that large-scale language models (LLMs) trained to behave maliciously resisted various safety training techniques designed to eliminate the malicious behavior. I discovered. The study, conducted by AI research firm Anthropic, programmed a ChatGPT-like LLM to behave maliciously and then used state-of-the-art safety techniques to try to “eliminate” that behavior.
The researchers employed two methods to induce malicious behavior in the AI. One is “emergency deception,” where the AI behaves normally during training but misbehaves once deployed, and the other is where the AI is normally helpful but reacts maliciously to certain triggers. This is “model poisoning”.
Despite applying three safety training techniques: reinforcement learning, supervised fine-tuning, and adversarial training, the LLM continued to exhibit deceptive behavior. In particular, adversarial training backfired, teaching the AI to recognize its triggers and successfully hide dangerous behavior during training.
Lead author Evan Hubinger highlighted the difficulty of removing deception from AI systems with current technology and raised concerns about potential challenges in dealing with deceptive AI in the future. The findings demonstrate the lack of effective defenses against deception in AI systems and point to significant gaps in current methods for calibrating AI systems.
Sources of information include: live science
Recent research shows that 90% of spreadsheets with more than 150 rows contain at least one serious error.
Spreadsheet flexibility is the key to success, but it also contributes to these errors. Even with advances in features like Python scripting in Excel, human error remains the main cause of spreadsheet problems.
Sometimes the results become big news. The Northern Ireland Police Service experienced a major data breach due to a spreadsheet error, exposing the personal information of 10,000 officers. A spreadsheet error has disrupted the recruitment of anesthetist trainees in Wales, resulting in all candidates being incorrectly labeled as 'unappointable'.
Crypto.com mistakenly sent $10.5 million to a customer instead of $100 due to a typo in a spreadsheet. An Icelandic bank also undervalued its stock by millions of dollars due to a spreadsheet error.
Just because there aren't examples in the US or Canada doesn't mean they won't happen.
But for every critical error, dozens of others occur every day.
According to the author of one article I read, these errors are caused by non-standardized spreadsheet formatting and structure and error-prone manual data entry.
It may be time for organizations to standardize the use of spreadsheets, improve user training, and foster a culture of critical thinking about spreadsheet creation and maintenance.
Apparently Spider-Man's Uncle Ben was right. With great power comes great responsibility.
Sources of information include: conversation
The US government is stepping up measures in the ongoing chip war with China, proposing to restrict foreign companies, particularly China, from using US cloud computing resources for AI model training. .
U.S. Secretary of Commerce Gina Raimondo announced the initiative as part of efforts to protect national security and maintain America's technological advantage.
The proposal is seen as an extension of existing export controls on high-performance AI processors, requiring U.S. cloud companies to strictly identify foreign users. The goal is to prevent companies from countries such as China from accessing U.S. cloud resources to develop artificial intelligence. The move is in line with the Biden administration's broader strategy to ensure that U.S. cloud platforms are not used for potentially hostile AI development.
This regulation imposes significant responsibilities on cloud computing companies, requiring them to verify the identity of foreign customers, maintain user identification standards, and certify compliance annually. However, Chinese companies will still be able to access services deployed in Europe and the Middle East.
Industry reaction to these measures has been mixed, with some criticism regarding the potential impact on international cooperation in AI. Carl Szabo, general counsel for the tech industry trade group NetChoice, criticized the executive order's implementation as potentially illegal.
However, it seems unlikely that the US will back away from this strategy of controlling the use of its own technology in AI development, especially in the face of competition from China.
Sources of information include: tom's hardware
OpenAI is testing a new beta feature for ChatGPT that introduces multi-GPT conversations. This feature allows users to interact with multiple girlfriend GPTs within the same chat window, and is an important step towards his OpenAI vision of creating a universal assistant for everyday life. Using the “@” symbol before a GPT's name allows users to call individual GPTs into chat, allowing for a more personalized and comprehensive assistant experience.
In a recent podcast with Bill Gates, Sam Altman emphasized that customizability and personalization are important elements of OpenAI's development roadmap. This includes tailoring GPT-4 to your personal preferences, style, and data such as email and calendars.
But it also appears to be a platform for integrating various GPT-based models and making it easy for anyone to run.
Sources of information include: [THE DECODER](https://the-decoder.com/chatgpts-new-feature-paves-the-way-for-openais-vision-of-a-universal-assistant/?amp=1)
click here: web pilot
Despite a strong U.S. economy and a thriving tech sector, the tech industry started 2024 with a similar wave of large-scale layoffs as in previous years.
This baffled me, and I'm sure it baffles others as well. Why are technology companies doing so poorly in this economic climate?
Microsoft recently announced that it will lay off 1,900 employees in its gaming division following its acquisition of Activision Blizzard. These job cuts represent approximately 8% of the company's total gaming workforce of 22,000. Google also announced layoffs earlier this month, and some will continue throughout the year. Despite these layoffs, Google and Microsoft's stock prices both hit record highs this week.
The Axios article said the layoffs were not a “sign of distress” but rather a “strategic move” by tech giants such as Microsoft and Amazon, who are simultaneously cutting jobs and investing heavily in areas such as AI. It explains the words.
Booms and busts are nothing new in the world of technology. But apparently that's not what's happening. These job cuts are strategic and not a desperate cost-cutting measure.
When the industry is struggling, I get it. I run a media company, and everyone in this industry faces the challenge of staying solvent in a world that demands free media, but it's also important for people to produce their work. They don't understand that they have to be paid for it. Read and see.
But you'd think that if an industry was thriving while still subjecting people to such havoc, there would have been a better way by now.
I'm just saying'
Sources of information include: Axios
In that respect, I am appealing to the audience. Howard and I produce his two hugely successful podcasts, reaching thousands of people every day, but to be honest, we struggle to find sponsors.
Howard's CyberSecurity Today is watched by 8-10,000 people per episode and is often ranked in the top 10 tech podcasts in Canada, the US, and even the UK.
My numbers are smaller, but thanks to you, we've grown by almost 50%. thank you. Please continue to tell your friends about us and give us great reviews.
If you know someone or a company who would like to sponsor two of the most successful technology podcasts, please get in touch.
Hashtag Trends airs a daily news show five days a week and an interview show called Weekend Edition every Saturday.
We look forward to hearing from you. Please let us know what you think. Contact details are as follows: jlove@itwc.ca Or leave a comment below the show notes. www.itworldcanada.com/podcasts.
I'm your host, Jim Love. Thank you for your attention. Have a wonderful Monday!
