Meta’s large language model gets leaked, DuckDuckGo launches its own AI-powered search engine, more than 60 per cent of data breaches caused by remote working.
 |
 |
 |
Welcome to Hashtag Trending for Tuesday, March 9th.
I’m your host Jim Love, CIO of IT World Canada and TechNewsDay in the US – here’s today’s top tech news stories.
Meta’s latest large language models, LLaMA has leaked online and is available for download, weeks after its release.
But Meta made its way online anyway, through a torrent leak.
Experts warned that similar technology could be used to automate the production of large amounts of fake news, spam, phishing emails, disinformation, incitement and more.
Meta, in fact, said last week that : “There is still more research that needs to be done to address the risks of bias, toxic comments, and hallucinations in large language models,” adding that it is releasing the model on a case-by-case basis to prevent misuse.
However, shortly after its release, someone on 4Chan posted details on how to obtain the whole model via peer-to-peer file sharing, and eventually instructions on how to download it all were published on GitHub.
The Register notes that the copies available on GitHub do appear to be legit. An AI engineer who wrote up the download instructions on Microsoft’s code-sharing site, showed the news site screenshots of him successfully generating text from the model.
While you may be strongly tempted to play with the leaked model, generally, only organizations that have access to piles of GPUs and other infrastructure are in a position to build, tweak, and test them. But, according to Meta’s AI researchers, LLaMA is built to be smaller and more compact and thus more accessible to academics and developers with smaller computational resources.
However, LLaMA does require hundreds of gigabytes of storage, but mostly a fair bit of knowledge to get the model up and running. Repurposing it for more nefarious purposes will take further technical expertise.
Meta said it will continue to share the model with selected researchers only despite the leak.
Source: The Register
The Chinese government has the ability to control the software on millions of devices through TikTok, FBI director Chris Wray revealed to the Senate Intelligence Committee.
Senator Marco Rubio asked, “Could they use it to drive narratives, like to divide Americans against each other?”
Wray replied, “Not only would that be possible, but we are not sure that we would see many of the outward signs of it happening, if it was happening”
He added, “something that’s very sacred in our country – the difference between the private sector and the public sector – that’s a line that is non-existent in the Chinese Communist Party.”
The Chinese government has a 1 per cent stake in one of ByteDance’s domestic subsidiaries as well as one of three seats on the subsidiary’s board.
TikTok, however, said it operates independently and protects U.S. data through an alliance with Oracle. Concerns about the company also included the firm’s previous violations of children’s privacy.
Source: Axios
However, if lawmakers do end up banning the China-based social media platform, Snap, Meta and Google would be the biggest beneficiaries, with billions of dollars of TikTok’s ad revenue up for grabs, according to research firm CFRA.
Senior equity analyst Angelo Zino at CFRA said, “Shares of select social media companies could have room to rise on a potential TikTok ban that could be gaining traction from Congress,”
Zino explained how “a TikTok ban would move the needle more for SNAP than others” given the similar demographic base using the platforms, adding that growth at YouTube could also be accelerated as the higher engagement levels and user base would be attractive for advertisers.
As a matter of fact, Snap shares increased 9.5 per cent on Monday after Senator Mark Warner of Virginia told Fox News he would introduce a bipartisan bill allowing the government to ban foreign technology from China.
The research firm, however, says that while the probability TikTok will be banned is still low, it has gone up significantly over the last six months.
Source: Insider
According to Fortinet’s new Work-from-Anywhere Global study, more than 60 per cent of companies had a data breach caused by vulnerabilities of remote working over the past 2 to 3 years.
570 companies were surveyed as part of the study. Out of which 60 per cent are accommodating employees working from home while 55 per cent are embracing a hybrid work strategy.
The high rate of breaches highlights that remote working hasn’t just introduced theoretical risks, but exposures that are being actively exploited by threat actors to gain access to sensitive information.
Organizations also cited a lack of cybersecurity training for hybrid workers as a main concern as well as how to extend corporate security to home offices and remote locations.
Organizations are also finding it difficult to enforce zero trust access controls and deploy patches to devices given the limited visibility over user’s home environments have limited visibility over user’s home environments, branch offices and off-site locations.
While most organizations have not reduced the risks associated with remote working, it did show that CISOs and security leaders are investing in new technologies to increase their cyber resilience.
For instance, respondents cited network access control, antivirus solutions, multi factor authentication (MFA) and cloud security solutions such as cloud access security broker (CASB) as the most important tools for securing remote work environments in the future.
Source: VentureBeat
The search engine that prides itself on protecting searchers’ privacy and avoiding filter bubbles now wears the medal of artificial intelligence.
DuckDuckGo launched a beta version of an AI search tool powered by ChatGPT called DuckAssist. The new search engine sources information from Wikipedia and related sources like the Encyclopedia Britannica to generate quick conversational responses to some questions.
The company said it will limit the data source to Wikipedia to avoid the misbehaviours that users have seen with other AI tools.
Unlike other ChatGPT-powered tools we have seen flooding since the start of this year, DuckAssist is not a chatbot, instead, it suggests an automatic answer when it recognizes a search term it can answer. You are neither forced to use the AI search tool, rather you’ll see a magic wand icon with an “ask me” button in your search results when an AI-powered response is available.
Plus, the company says it’s maintaining privacy as its focus by keeping users’ search, and browsing history anonymous in all data shared with OpenAI, and Anthropic, the company’s partners on this project.
The tool is free and now available on the DuckDuckGo web browsing apps for phones and computers as well as the company’s browser extension.
Source: GIZMODO
Those are the top tech news stories for today
Links to these stories can be found in the article posted on itworldcanada.com/podcasts. You can also find more great stories and more in-depth coverage in itworldcanada.com or in the US on technewsday.com
Hashtag Trending goes to air five days a week with a daily newscast and we have a special weekend edition with an interview featuring an expert in some aspect of technology that is making the news.
We’re always happy to hear from you, you can find me on LinkedIn, Mastodon, Twitter or just leave a comment under the article for this podcast at ITWorldCanada.com/podcasts.
I’m your host Jim Love – Have a great Thursday!