Dan Costa

Richard Feynman on the difference between knowing the name of something and knowing something. Reminded me of what Yan LeCunn said about LLM's intelligence #AI and real intelligence. 

30

2 Comments

Through a strategic initiative with Prefix.dev announced Wednesday, Anaconda will improve the speed and efficiency of its open-source ‘conda-build’ development tool used by Anaconda customers and the open-source community. https://okt.to/MwCdfR

2

Smart commentary here for journalists—and all content creators—on the type of content that will win and lose in a world of AI search. But as someone who is a connoisseur of the explainer form, I suspect that headline may be overstated, and that only a certain type of explainer is dead—explainers that AI can easily duplicate. AI can’t call two or three experts with original insights and put together a second-day story explaining what caused a ship to hit a bridge or made that bridge vulnerable to collapse. It can’t mine the brain of an expert in the laws of war who can offer accurate insight into when a strike on the open seas is illegal. I’m curious about your thoughts—what types of content will be valuable in a world of AI search and what won’t?

28

4 Comments

Alrighty, I've been reading many of these, so my turn to chime in. I worked at WTOP News for nine years, filed stories for CBS Radio as a reporter, used their content when I was a producer. I'm disappointed but not surprised. People consume news on social media through their phones. People also don't need traffic reports as much due to Waze and Google Maps. So the question becomes: what makes a radio outlet unique with content that's not available elsewhere. WTOP News and other local affiliates are in good shape, in my opinion, despite the loss of WCBS 880 in NYC. You can't get hyperlocal coverage of Montgomery County, MD or Fairfax County VA without WTOP. It doesn't exist. Julie Ziegler preaches the same message Jim Farley did years ago: local, local, local. CBS Radio News didn't have that on a national level. Now what they brought that the market may miss are the quality stories from the stringers, like I was, across the country. An important story in Detroit that matters in Philadelphia, but the national news doesn't cover. That's what we'll miss. On Bari Weiss: I don't wish her ill will. If CBS News Radio was operating deep losses, then it's a necessary business decision while painful. However, all her slashing and slashing of costs will amount to a bunch of nothing UNLESS she replaces it with new MONEY-MAKING content. Anyone can cut costs with a scalpel or sledgehammer. What's her idea to make CBS News profitable? We wait and see...

11

6 Comments

#RSAC 2025 is expected to continue the focus on AI from the conferences of prior years, but with a new emphasis this time on the emergence of agentic capabilities. Along with GenAI and AI agents, other key topics will include: - The evolution of security operations - Protection against intensifying attacks targeting the cloud and email inboxes - The ongoing push for consolidation and “platformization” and - The uncertain economic outlook Here’s where you’ll find all of CRN’s news coverage from RSA Conference 2025 ⬇️

6

Elsevier has collaborated with the Institute of Scientific and Technical Information of China and other international publishers to release the “Guideline on the Boundaries of AIGC (AI-generated content) Usage in Academic Publishing 2.0,” which promotes transparency and accountability in AI use in research. This initiative aligns with global discussions on AI’s impact on research integrity, as highlighted at the Third Sino-Swiss Research Integrity Workshop. The workshop addressed the challenges and opportunities AI presents in academic publishing, emphasizing the need for clear standards and responsible AI application. #Elsevier #Research #Publishing #ResponsibleAI

12

1 Comment

This is in theory a very nerdy question, but one that has pretty big implications for assessing whether AI is going to do what we want it to or not. https://lnkd.in/ewH35Cbk

4

This week's post: On a fascinating idea to try to ground LLMs in — if not truth, at least in sourcing — and what that could mean for journalism and public interesting information. Plus, structured journalism. https://lnkd.in/epjz3bAv

17

"Critics have said that newspapers that use AI tools risk exposing readers to low-quality reporting and misinformation, contributing to a rising mistrust of journalism." Facts matter. Fact-checking is critical. The machine is not ready. It's a tool - treat it as such. https://wapo.st/3H1EPMd #storiesthatmatter #journalismmatters #factsmatter #AIisatool

15

The latest from Future.News --> "Like it or not, AI is going to fundamentally change the way in which we work, the way in which news is created, and the way in which audiences engage with news," Associated Press CEO Daisy Veerasingham tells Barron's https://lnkd.in/eK8UGr2y

4

Critique of Substack’s Approach to Email Delivery Enforcement The message from Substack Standards & Enforcement, while framed politely, reflects a concerning and overly punitive approach to handling email delivery metrics, particularly for publishers who rely on the platform to build and engage with their communities. First and foremost, disabling publishing abilities without prior warning or a transparent, data-driven explanation undermines the trust between creators and the platform. The vague reference to “poor email delivery indicators,” such as low open rates or spam complaints, is insufficient. Publishers deserve access to detailed analytics before such drastic action is taken. Without clarity or specific thresholds, the criteria appear arbitrary and opaque. Furthermore, placing the burden entirely on the creator to “prune” their list through a re-confirmation process penalizes legitimate audience-building efforts. Many imported lists consist of genuine followers collected from personal websites, events, or past platforms. Forcing re-verification not only risks losing engaged readers but also ignores the nuances of user behavior—open rates and spam complaints can vary due to numerous external factors, including email client algorithms or misleading spam filters. The enforcement email also lacks empathy and support for content creators, offering no collaborative solution or opportunity to appeal. A more constructive approach would involve education, tailored feedback, and system support (e.g., A/B testing subject lines or improved deliverability tips) rather than immediate suspension of core functionality. Lastly, it is contradictory for a platform built on creator empowerment and independence to impose such heavy-handed controls without due process or proportional response. Substack should rethink this policy and replace enforcement through disruption with engagement through support. Empowering creators to improve deliverability, rather than halting their work, would reflect a more ethical and sustainable strategy aligned with Substack’s supposed mission.

9

2 Comments

"Speed is the disease of the digital age. In today’s newsrooms, pressure to publish stories at an increasingly fast rate often means that reporters are able to uncover only basic information and complete a surface-level analysis. Twenty years ago media experts were already warning that working at high speeds encourages journalists “to fall back on well-worn themes and observations—interpretive clichés.” In today’s world of digital media, this effect has been compounded." https://lnkd.in/g7gGAzKS

29

1 Comment

MA Senate passes comprehensive privacy bill: A company complying with Maryland’s data minimization standard would be in compliance with a similar measure proposed in a Massachusetts comprehensive privacy bill that’s moving quickly toward passage, said Massachusetts Sen. Michael Moore (D) on the floor Thursday. However, Moore also said he’s fine with Massachusetts being an “outlier” among the 20 states with privacy laws. Read in-depth coverage of yesterday's vote in Privacy Daily here: https://lnkd.in/d2VdRHmT

13

2 Comments

Will AI Reshape the Value Proposition of Higher Ed? A new Chronicle Intelligence research brief, produced with support from CollegeVine, explores how AI is pushing higher ed to confront fundamental questions about value, work-force readiness, and institutional strategy. 🔹 Eight in 10 administrators and faculty members say AI is forcing them to reconsider the value colleges offer students. 🔹 Yet only one in four institutions has taken concrete action to adapt. 🔹 Faculty and administrators remain deeply divided over how far — and how fast — to move. The report examines where institutions see opportunity, what barriers are holding them back, and how they’re balancing innovation with tradition as AI transforms teaching, learning, and operations. #HigherEd #AI #EdTech #InstitutionalStrategy https://lnkd.in/e34bPNFv

8

Lately, when I tell people that I am a writer, the conversation inevitably turns to artificial intelligence. People want to know if AI will replace journalists and take over everything from writing newspapers to books. I don't think it will, and here's why. It’s what I write about this week https://lnkd.in/e2F-a2Aa

2

A housing reporter at CT Mirror in Connecticut had an 8-hour legislative committee meeting to cover. She got through it in 2 hours by reading the AI-generated transcript instead of listening to the recording. The tool: Legitalk, built by CT Mirror's AI data reporter Angela Eichhorst. Launched February 4. What I found interesting: Eichhorst didn't start by asking "What can AI do?" She asked 15 colleagues: "What annoys you at work?" She got 85 pain points back. Most weren't technology problems. One of the result of asking about pain points: A tool that auto-transcribes Connecticut's 30+ legislative committee hearings and lets reporters jump directly to the moments that matter. No more scrubbing through 10-hour videos. But the real story is in the details - how she tuned it based on reporter feedback within days of launch, why off-the-shelf tools keep failing newsrooms, and what makes this replicable (or not). Also: the time she had to rephrase prompts multiple times because OpenAI kept blocking her for searching "explicit language" in a 12-hour legislative video. Full story in this week's News Machines. https://lnkd.in/gzH39pET #AIinJournalism #MediaInnovation #NewsroomTools

2

A thoughtful (paraphrased) comment from Mohammad Hosseini that has particularly stuck with me following National Information Standards Organization's webinar on Research Integrity earlier today: The use of generative AI is a double edged sword. On one hand it can help save time during the research process, but on the other hand it adds time to the process as every claim, quote, and citation put out by generative AI tools neeeds to be double checked against the original source for accuracy. This verification process often costs more time than the initial prompt saved. Responsible and transparent use of generative AI can greatly benefit the research process, but ultimately it is the researchers' responsibility to ensure that the output is correct, reliable, and free from hallucinations and incorrect interpretations of the source data.

22

1 Comment