h
LightNews LightNews
banner
angelnumbers.bsky.social
h
@angelnumbers.bsky.social
26 followers 87 following 6 posts
angelnumbers.carrd.co
Posts Replies Media Videos
angelnumbers.bsky.social
Someone needs to make an american version of Mad Max about teamsters
December 10, 2025 at 6:54 AM
Reposted by h
nicetopatrick.bsky.social
food takes are what got jean valjean into trouble
November 25, 2025 at 10:20 PM
Reposted by h
depthsofwikipedia.bsky.social
huge news I just found a folder of these that I forgot i had!!!!!!!
November 25, 2025 at 11:50 PM
Reposted by h
reckless.bsky.social
I’ve been running around asking tech execs and academics if language was the same as intelligence for over a year now - and, well, it isn’t. @benjaminjriley.bsky.social explains how the bubble is built on ignoring cutting-edge research into the science of thought www.theverge.com/ai-artificia...
November 25, 2025 at 1:54 PM
angelnumbers.bsky.social
Claude said it was citing research on the topic I was asking it about, but the sources were continually wrong
It then just gave up and told me what I wanted to hear, which is that it should stop pretending to know how to read :-(
November 25, 2025 at 6:41 PM
Reposted by h
ketanjoshi.co
Deeply absurd. This Google PDF published on a blog (arxiv, not peer reviewed) claims an LLM is "PhD level" but in most cases the MAJORITY of reference URLs were invalid or inaccessible.

A PhD sitting down and just fabricating >50% of sources = career ending

arxiv.org/abs/2511.11597
CLINB: A Climate Intelligence Benchmark for Foundational Models Michelle Chen Huebscher1 , Katharine Mach2 , Aleksandar Stanić1 , Markus Leippold1,3, Ben Gaiarin1 , Zeke Hausfather4 , Elisa Rawat , Erich Fischer5 , Massimiliano Ciaramita1 , Joeri Rogelj6 , Christian Buck1 , Lierni Sestorain Saralegui1 and Reto Knutti5 1Google DeepMind, 2University of Miami, 3University of Zurich, 4Stripe, 5ETH Zurich, 6 Imperial College London Evaluating how Large Language Models (LLMs) handle complex, specialized knowledge remains a critical challenge. We address this through the lens of climate change by introducing CLINB, a benchmark that assesses models on open-ended, grounded, multimodal question answering tasks with clear requirements for knowledge quality and evidential support. CLINB relies on a dataset of real users’ questions and evaluation rubrics curated by leading climate scientists. We implement and validate a model-based evaluation process and evaluate several frontier models. Our findings reveal a critical dichotomy. Frontier models demonstrate remarkable knowledge synthesis capabilities, often exhibiting PhD-level understanding and presentation quality. They outperform “hybrid" answers curated by domain experts assisted by weaker models. However, this performance is countered by failures in grounding. The quality of evidence varies, with substantial hallucination rates for references and images. We argue that bridging this gap between knowledge synthesis and verifiable attribution is essential for the deployment of AI in scientific workflows and that reliable, interpretable benchmarks like CLINB are needed to progress towards building trustworthy AI systems. Total Reference URLs Generated claude-opus-4-1 claude-sonnet-4 gpt-5 hybrid gemini-2.5-pro gemini-2.5-flash o3 0.0 0.2 0.4 0.6 0.8 1.0 Proportion Reference URL Status hybrid gemini-2.5-pro claude-opus-4-1 o3 gemini-2.5-flash claude-sonnet-4 0 200 400 600 800 1000 Count of URLs Total Image URLs Generated hybrid claude-opus-4-1 gemini-2.5-flash gemini-2.5-pro claude-sonnet-4 o3 0.0 0.2 0.4 0.6 0.8 1.0 Proportion Image URL Status Status OK INACCESSIBLE_CONTENT INVALID_URL ERROR Figure 3 | Number of reference (top), and image (bottom), URLs and their status. Ablations We perform several ablation studies with the autorater (Table 4). Notably, removing the question-specific rubrics from the prompt changes the results only in the bottom half, with the Hybrid answers overtaken by Gemini 2.5 Flash and Claude Sonnet 4. This suggests that the additional resolution provided by the rubrics applies primarily to the kind of responses used to develop the rubrics. Or, in other words, that rubrics are far from complete. Hence, it is important that rubrics adapt to new data as better models become availab A New Expert-Grounded Benchmark for Scientific AI We introduce CLINB, a benchmark for modelbased evaluation of frontier models on complex, multimodal scientific communication. Its core is a new dataset of real-world climate questions paired with data-driven, question-specific evaluation rubrics, curated and validated by leading climate scientists through a novel three-phase, human-in-the-loop process.2 PhD-Level Synthesis vs. Attribution Failures Frontier models demonstrate remarkable knowledge synthesis, often exhibiting a PhD-level understanding. However, this performance masks a critical inadequacy in grounding. We report substantial hallucination rates for references (10% to 25%) and even more failures for images (50% to 80% in certain settings), exposing a major gap between synthesis and verifiable attribution. Insights into Human-AI Collaboration Dynamics Autonomous frontier models surpass ’hybrid’ answers (curated by experts using weaker AI assistance), revealing the assisting model’s capability—not human oversight—as the primary bottleneck. Counter-intuitively, highly motivated non-specialists (our ’Advocates’) who deeply engage with AI tools can produce higher-quality answers than domain experts who engage less with AI during answer curation. A Validated Methodology for Scalable Oversight We validate a rigorous, rubric-based autorater. Ablation studies demonstrate that structured prompts and automated evidence-checking are essential for mitigating inherent LLM judge biases. This process is hampered by inaccessible sources (up to 50%). Furthermore, we identify evaluation challenges, including model familiarity bias in human raters and the limitations of rubrics to generalize across models.
November 24, 2025 at 7:36 PM
Reposted by h
riibrego.bsky.social
"there are no third spaces anymore" wrong. blast furnace
November 25, 2025 at 12:59 AM
Reposted by h
moreperfectunion.bsky.social
Amazon is now integrating betting platform FanDuel directly into Amazon Prime.

Bettors will be able to link their FanDuel accounts to their Prime profiles and see how their NBA wagers are playing out in real time, according to Amazon’s latest announcement.
September 30, 2025 at 7:41 PM
angelnumbers.bsky.social
Maybe the world is falling apart but thanks to modern technology I can go to vr raves on my MacBook and dance shyly in the corner
September 28, 2025 at 7:33 AM
Reposted by h
tuckfriendly.bsky.social
None of the videos were in black and white, so I figured we were ok.
June 7, 2025 at 5:09 PM
Reposted by h
vortexegg.com
Insecurity theater
collie.bsky.social collie.bsky.social @collie.bsky.social · Jun 3
I cannot get over these cops wearing night vision to arrest a dude on a Southwest Airlines flight in broad daylight.
June 3, 2025 at 3:52 PM
Reposted by h
dorwright.bsky.social
LRT; i think about recycling whenever discussion of cybersecurity comes up. We've been successfully gaslight into thinking its an issue of personal responsibility but the problems are entirely the result of corporate cost-cutting and accountability dodging.
May 28, 2025 at 2:46 PM
Reposted by h
tofupixel.bsky.social
some thing
May 28, 2025 at 10:12 PM
Reposted by h
northerniion.bsky.social
extremely important PSA
February 11, 2025 at 4:30 AM
Reposted by h
jonbois.bsky.social
this place is the mines of moria for people who say “superb owl.” tread with the utmost care
February 9, 2025 at 4:53 PM
angelnumbers.bsky.social
February 8, 2025 at 1:37 AM
Reposted by h
jamellebouie.net
is it good or bad that a billionaire oligarch with close ties to foreign dictators is currently downloading the most sensitive information the us federal government has about its citizens to his own private servers?
February 2, 2025 at 6:36 AM
Reposted by h
mothcub.bsky.social
Thank you FKA Twigs for my life.
A drawing of a woman with a strange, toad like expression listening to the album "Eusexua" by FKA Twigs. The drawing is labelled "Tue 28 Jan '25" and text reads: "I am feeling it".
February 2, 2025 at 1:11 PM
Reposted by h
uhwooguh.bsky.social
Did Baby ChatGPT just rizz up Livvy Dunne??? DeepSeek could be the new Rizz King
January 28, 2025 at 6:11 PM
Reposted by h
lackingsaint.bsky.social
wikipedia honestly gotta be top 5 human creations in the past thousand years
January 22, 2025 at 3:11 AM
Reposted by h
goldentreebe.bsky.social
January 21, 2025 at 1:52 AM
Reposted by h
jonbois.bsky.social
those are my drones. stop talking about them. stop looking at them. stop pointing at them. they're none of your business. they have nothing to do with you. leave me and my drones alone
December 16, 2024 at 10:57 PM
Reposted by h
humanoidhistory.bsky.social
Dear Santa letter from the Mathis News in Texas, December 21, 1978.
December 10, 2024 at 7:41 PM
Reposted by h
bigjoel.bsky.social
Seriously don’t ever elect me president because if you EVER throw my boy in the slammer, he’s gonna get an early Christmas gift from papa. He’s gonna get his freedom
December 2, 2024 at 5:29 AM
Reposted by h
bombsfall.bsky.social
The early glimpses of cinematic talent that Coppola showed in The Godfather really come into full view in his true masterpiece, Bram Stoker's Dracula
December 2, 2024 at 5:34 AM