Skip to main content

DeepSeek limits new accounts amid cyberattack

3 months 2 weeks ago
Chinese AI startup grapples with consequences of sudden popularity

Updated  China's DeepSeek, which shook up American AI makers with the debut of its V3 and reasoning-capable R1 LLM families, has limited new signups to its web-based interface to its models due to what's said to be an ongoing cyberattack.…

Thomas Claburn

Anthropic Builds RAG Directly Into Claude Models With New Citations API

3 months 2 weeks ago
An anonymous reader quotes a report from Ars Technica: On Thursday, Anthropic announced Citations, a new API feature that helps Claude models avoid confabulations (also called hallucinations) by linking their responses directly to source documents. The feature lets developers add documents to Claude's context window, enabling the model to automatically cite specific passages it uses to generate answers. "When Citations is enabled, the API processes user-provided source documents (PDF documents and plaintext files) by chunking them into sentences," Anthropic says. "These chunked sentences, along with user-provided context, are then passed to the model with the user's query." The company describes several potential uses for Citations, including summarizing case files with source-linked key points, answering questions across financial documents with traced references, and powering support systems that cite specific product documentation. In its own internal testing, the company says that the feature improved recall accuracy by up to 15 percent compared to custom citation implementations created by users within prompts. While a 15 percent improvement in accurate recall doesn't sound like much, the new feature still attracted interest from AI researchers like Simon Willison because of its fundamental integration of Retrieval Augmented Generation (RAG) techniques. In a detailed post on his blog, Willison explained why citation features are important. "The core of the Retrieval Augmented Generation (RAG) pattern is to take a user's question, retrieve portions of documents that might be relevant to that question and then answer the question by including those text fragments in the context provided to the LLM," he writes. "This usually works well, but there is still a risk that the model may answer based on other information from its training data (sometimes OK) or hallucinate entirely incorrect details (definitely bad)." Willison notes that while citing sources helps verify accuracy, building a system that does it well "can be quite tricky," but Citations appears to be a step in the right direction by building RAG capability directly into the model. Anthropic's Alex Albert clarifies that Claude has been trained to cite sources for a while now. What's new with Citations is that "we are exposing this ability to devs." He continued: "To use Citations, users can pass a new 'citations [...]' parameter on any document type they send through the API."

Read more of this story at Slashdot.

BeauHD

Facebook Flags Linux Topics As 'Cybersecurity Threats'

3 months 2 weeks ago
Facebook has banned posts mentioning Linux-related topics, with the popular Linux news and discussion site, DistroWatch, at the center of the controversy. Tom's Hardware reports: A post on the site claims, "Facebook's internal policy makers decided that Linux is malware and labeled groups associated with Linux as being 'cybersecurity threats.' We tried to post some blurb about distrowatch.com on Facebook and can confirm that it was barred with a message citing Community Standards. DistroWatch says that the Facebook ban took effect on January 19. Readers have reported difficulty posting links to the site on this social media platform. Moreover, some have told DistroWatch that their Facebook accounts have been locked or limited after sharing posts mentioning Linux topics. If you're wondering if there might be something specific to DistroWatch.com, something on the site that the owners/operators perhaps don't even know about, for example, then it seems pretty safe to rule out such a possibility. Reports show that "multiple groups associated with Linux and Linux discussions have either been shut down or had many of their posts removed." However, we tested a few other Facebook posts with mentions of Linux, and they didn't get blocked immediately. Copenhagen-hosted DistroWatch says it has tried to appeal against the Community Standards-triggered ban. However, they say that a Facebook representative said that Linux topics would remain on the cybersecurity filter. The DistroWatch writer subsequently got their Facebook account locked... DistroWatch points out the irony at play here: "Facebook runs much of its infrastructure on Linux and often posts job ads looking for Linux developers."

Read more of this story at Slashdot.

BeauHD

2025 Will Likely Be Another Brutal Year of Failed Startups, Data Suggests

3 months 2 weeks ago
An anonymous reader quotes a report from TechCrunch: TechCrunch gathered data from several sources and found similar trends. In 2024, 966 startups shut down, compared to 769 in 2023, according to Carta. That's a 25.6% increase. One note on methodology: Those numbers are for U.S.-based companies that were Carta customers and left Carta due to bankruptcy or dissolution. There are likely other shutdowns that wouldn't be accounted for through Carta, estimates Peter Walker, Carta's head of insights. [...] Meanwhile, AngelList found that 2024 saw 364 startup winddowns, compared to 233 in 2023. That's a 56.2% jump. However, AngelList CEO Avlok Kohli has a fairly optimistic take, noting that winddowns "are still very low relative to the number of companies that were funded across both years." Layoffs.fyi found a contradicting trend: 85 tech companies shut down in 2024, compared to 109 in 2023 and 58 in 2022. But as founder Roger Lee acknowledges, that data only includes publicly reported shutdowns "and therefore represents an underestimate." Of those 2024 tech shutdowns, 81% were startups, while the rest were either public companies or previously acquired companies that were later shut down by their parent organizations. So many companies got funded in 2020 and 2021 at heated valuations with famously thin diligence, that it's only logical that up to three years later, an increasing number couldn't raise more cash to fund their operations. Taking investment at too high of a valuation increases the risk such that investors won't want to invest more unless business is growing extremely well. [...] Looking ahead, Walker also expects we'll continue to see more shutdowns in the first half of 2025, and then a gradual decline for the rest of the year. That projection is based mostly on a time-lag estimate from the peak of funding, which he estimates was the first quarter of 2022 in most stages. So by the first quarter of 2025, "most companies will have either found a new path forward or had to make this difficult choice." "Tech zombies and a startup graveyard will continue to make headlines," said Dori Yona, CEO and co-founder of SimpleClosure. "Despite the crop of new investments, there are a lot of companies that have raised at high valuations and without enough revenue."

Read more of this story at Slashdot.

BeauHD