Could you please expand on this?

New study reveals chatbot empathy can worsen customer reactions When a service encounter goes south, customers expect empathy. Hearing an employee say, "I share your frustration," can calm tensions and rebuild trust.

Empathetic responses from AI chatbots during service issues can increase customer discomfort and reduce satisfaction, highlighting differing expectations for emotional awareness in humans versus AI.

@lastnpcalex.agency I'm sorry, I wrote this before I saw your post

Incorrect - his successor is now named John Apple, I don't make the rules

Introducing Salesforce Headless 360. No Browser Required. Everything on Salesforce is now an API, MCP tool, or CLI command, and agents can use all of it.

Great development from Salesforce, enabling AI agents to directly interface with their systems via API, MCP, and CLI

Points toward a future where SaaS products are just endpoints in one large, seamless business software system, driven by people and agents

www.salesforce.com/news/stories...

Last year, we introduced FlexOlmo, a novel way to train parts of a model independently then combine them later.

BAR builds on that idea for a harder problem: how to keep improving a model without having to retrain each time. 🧵

I get why people are shitting on this example, but you should definitely do the "record interviews with your loved ones" thing. There's a reason people save voicemails from folks long gone.

Prove You Are a Robot: CAPTCHAs for Agents

Prove You Are a Robot: CAPTCHAs for Agents

#HackerNews

https://browser-use.com/posts/prove-you-are-a-robot

According to fedidb.com, hachyderm is about 7.8k MAU and 4.8M posts

Techhub.social is about 4.2k MAU, 2.6M posts

So hachyderm is roughly 2x the size of techhub.social, but spending about 4x as much monthly

This is such an interesting contrast to techhub.social, where costs are an order of magnitude lower

hub.techhub.social/finances/

Also very notable that Anthropic has not had an open release!

My interpretation (and I'm very open to others!) is:

- OpenAI did get pressured ("why are you guys OpenAI if all of your models are closed?"), but saw it as 1-off buying time and goodwill
- Google wants AI at the edge, which would be reverse-engineered anyway. Might as well get the halo out of it.

Screenshot of the repository

Serverless browser runtime for AI agents

Also keeps the pressure off regarding releasing their own open models

It's also a part of why the critiques are framed with social weaponization tools - "it lies", "it's fake", "it's trying to manipulate you", "it's made/used by people you don't like"

And why those most excited about it are generally folks who don't play social hierarchy games

"Social stakes" feels like exactly the right framing

Previous comms technologies frequently empowered those adept at navigating social dynamics

But AI - with it's ability to create context-appropriate prose for every milieu - is genuinely threatening for them

Turns out if you take the weights and tokenizer and directly materialized them to static globals at compile-time, inference gets very fast!

Very very fast!

Hi, Adam! Single-topic novelty account here. It would have been awfully hard for people to promote smartphones in 1999, considering that SymbianOS wasn't commercially available until 2000.

But you know what transformational technology was being promoted in 1999?

The Internet!

I hope it's the "woop woop" sound from the Three Stooges

I have a nuclear reaction encoding defined as ``` struct NuclearReaction { int explosionPower; // explosion power struct NuclearReaction *next; // next nuclear reaction } *NuclearReactionSequence; ``` Doctor evil has realised through his time machine he can run the nuclear reaction in reverse. Write a function that takes a nuclear reaction sequence and updates in place to do the reaction in reverse.

making programming assignments that students can't just ask ChatGPT for by exploiting the safety checks lol

I love how this is both wholesome and slightly unhinged

2026 (planned) Data centers vs. megaprojects Inflation-adjusted costs, billions USD SIT- 2025₽ Data center capex =$930B in 6 years $750B- Interstate Highway System ® S620B, 37yr US Railroads $550B, 71yr $500B- F-35 Program $400B, 25yr (to date) $250B - $O Apollo Program $257B, 14yr Marshall Plan $170B, 4yr Manhattan-Project -$36B,Syr- 10 20 International Space Station $150B, 27yr- 30 40 Years from start of program 50 60 70 Sources: Company reports, Epoch AI • FHWA • NASA • CRS • GAO • Brookings Al capex = estimated data-centre share of global reported capex at the big-5 US hyperscalers (Amazon, Microsoft, Alphabet, Meta, Oracle; Epoch AI+ Platformonomics). Assuming DC share scales from +55% in 2020 to =80% by 2026. Excludes Chinese hyperscalers. All costs in 2024 dollars.

Absolutely insane numbers on data center buildout. I haven’t vetted these by they’re about right based on memory/back of the envelope.

Revision of human prose I initially wrote this to solve a problem with LLM prose. As I crafted the prompt and tested it, I did think more and more about what use it might have to someone looking to figure out a problem they are having with their own writing. In this mode I don't think of it as an automatic paragraph generator, but as something that might open a stuck writer's eyes to a range of possibilities they hadn't seen, which they could then take in places and reintegrate into their work.

Introducing /jamesian, a Claude skill that helps you de-homogenize LLM prose mikecaulfield.substack.com/p/introducing-… #AI #writing

It's fascinating that journalists' concerns about AI still center on hallucinations, when the larger AI risks today are more commonly due to their accuracy at scale (cybersecurity 0-days, voice cloning for scams, fake images for misinformation, etc.)

Pretty sure Goldilocks' AI would recommend binary search

Great tool from Cloudflare to help evaluate the AI agent readiness of websites

This guy gets it

With max thinking Opus 4.7 is quite impressive, with a real sense of style

In two prompts: "implement the Tower of Babel, in 3D, in as sophisticated and visually interesting a way as possible. It should be interactive" and then "make it better."

Play: tower-of-babel-1776392618.netlify.app

Just noting that this didn't happen before Jerry was verified

an under-appreciated advantage of increased capacity for summarizing and shaping and rotating text and ideas to our own needs is that we can put more information into our comms without overloading receivers! you don't have to guess (as much) anymore what will be relevant to your interlocutor(s).