GPT-4 Has Passed the Turing Test, Researchers Claim
ELIZA, a system pre-programmed with responses but with no large language model (LLM) or neural network architecture, was judged to be human just 22% of the time. GPT-3.5 scored 50% while the human participant scored 67%. “Machines can confabulate, mashing together plausible ex-post-facto justifications for things, as humans do,” Nell Watson, an AI researcher at the Institute of Electrical and Electronics Engineers (IEEE), told Live Science. “They can be subject to cognitive biases, bamboozled and manipulated, and are becoming increasingly deceptive. All these elements mean human-like foibles and quirks are being expressed in AI systems, which makes them more human-like than previous approaches that had little more than a list of canned responses.” Further reading: 1960s Chatbot ELIZA Beat OpenAI’s GPT-3.5 In a Recent Turing Test Study
Read more of this story at Slashdot.
Electricity Bills Forecasted To Climb With Summer Heat
To be sure, the EIA says that weather is “the main source of uncertainty” in its forecasts for folks’ utility bills. If this summer winds up being hotter than expected, households could wind up paying even more. Residential electricity use typically peaks in the summer for most of the US because of air conditioning. Extreme heat can even trigger power outages if demand suddenly rises too sharply. California, the Southwest, the Midwest, Texas, and New England are at “elevated risk” of electricity supply shortages during any extreme weather this summer, according to an assessment (PDF) by the North American Electric Reliability Corporation.
Read more of this story at Slashdot.
Ransomware Attackers Quickly Weaponize PHP Vulnerability With 9.8 Severity Rating
CVE-2024-4577 affects PHP only when it runs in a mode known as CGI, in which a web server parses HTTP requests and passes them to a PHP script for processing. Even when PHP isn’t set to CGI mode, however, the vulnerability may still be exploitable when PHP executables such as php.exe and php-cgi.exe are in directories that are accessible by the web server. This configuration is extremely rare, with the exception of the XAMPP platform, which uses it by default. An additional requirement appears to be that the Windows locale — used to personalize the OS to the local language of the user — must be set to either Chinese or Japanese. The critical vulnerability was published on June 6, along with a security patch. Within 24 hours, threat actors were exploiting it to install TellYouThePass, researchers from security firm Imperva reported Monday. The exploits executed code that used the mshta.exe Windows binary to run an HTML application file hosted on an attacker-controlled server. Use of the binary indicated an approach known as living off the land, in which attackers use native OS functionalities and tools in an attempt to blend in with normal, non-malicious activity.
In a post published Friday, Censys researchers said that the exploitation by the TellYouThePass gang started on June 7 and mirrored past incidents that opportunistically mass scan the Internet for vulnerable systems following a high-profile vulnerability and indiscriminately targeting any accessible server. The vast majority of the infected servers have IP addresses geolocated to China, Taiwan, Hong Kong, or Japan, likely stemming from the fact that Chinese and Japanese locales are the only ones confirmed to be vulnerable, Censys researchers said in an email. Since then, the number of infected sites — detected by observing the public-facing HTTP response serving an open directory listing showing the server’s filesystem, along with the distinctive file-naming convention of the ransom note — has fluctuated from a low of 670 on June 8 to a high of 1,800 on Monday. Censys researchers said in an email that they’re not entirely sure what’s causing the changing numbers.
Read more of this story at Slashdot.
A Growing Number of Americans Are Getting Their News From TikTok
Read more of this story at Slashdot.