- Forward Future Daily
- Posts
- š§āš Global AI Safety, Broken Benchmarks & Appleās Smart Glasses Play
š§āš Global AI Safety, Broken Benchmarks & Appleās Smart Glasses Play
Singapore pushes AI safety accord, SWE-Bench faces criticism, Apple readies smart glasses, Stripe expands AI and stablecoins, xAI scales up, LLMs still hallucinate.
Good morning, itās Friday. Singaporeās playing AI diplomat as global tensions rise, benchmarks are breaking under pressure as we struggle to measure real intelligence, and Appleās quietly plotting to strap even more tech to your face.
Plus, in todayās Forward Future Original, we explore how the second half of 2025 could mark AIās shift from passive assistant to active collaboratorādriven by massive context windows, agentic systems, and memory-based personalization.
Read on!
š¤ FRIDAY FACTS
What do Pope Francis and an AI-generated puffer jacket have in common?
Stick around to find out! š
šļø YOUR DAILY ROLLUP
Top Stories of the Day

š¶ļø Apple Developing Smart Glasses with & without AR
Apple is designing custom chips for two types of smart glasses: one focused on everyday smart features and another with full AR capabilities. The chip, adapted from Apple Watch tech, supports multiple cameras and may enter mass production by 2026 or 2027. These glasses could rival Meta's offerings. Apple is also updating chips for AirPods, Watch, and AI servers company-wide.
š³ Stripe Supercharges AI, Stablecoins, and Global Payments
Stripe has launched the first AI foundation model for payments and new stablecoin-powered accounts in 101 countries, aiming to streamline global money management. The companyās multicurrency and card issuing features help businesses avoid FX fees and spend stablecoins like cash. A fast-growing enterprise client base, including NVIDIA and PepsiCo, underscores its expanding influence.
š§® Musk's xAI Supercluster Hits 200K GPUs, Eyes 300 MW
Elon Muskās xAI Colossus supercomputer in Memphis is now fully operational, powered by 150 MW from the grid and 150 MW from Tesla Megapack batteries. The GPU count has doubled to 200,000, with plans to scale up to one million. Phase 2, launching this fall, will require 300 MWāenough to power 300,000 homes. Temporary gas turbines will be phased out as a new substation comes online.
šµāš« Popular LLMs Hallucinate Confidently, Phare Study Finds
Phare, a new multilingual benchmark, reveals leading LLMs often produce factually incorrect but authoritative-sounding responses, especially under misinformation prompts. The study shows that user framing, like confident tone, significantly increases hallucination rates. Models optimized for user satisfaction can prioritize plausibility over truth, posing real risks in high-stakes use cases.
Enjoying our newsletter? Forward it to a colleagueā
itās one of the best ways to support us.
šŗļø GEOPOLITICS
Singapore Brokers Global AI Safety Accord Amid US-China Tensions

In a rare display of unity, top AI researchers from the US, China, and Europe gathered in Singapore to endorse a joint blueprint for international cooperation on AI safety. The āSingapore Consensusā outlines shared research priorities for managing risks from frontier AI systemsāincluding controlling their behavior and developing safer model architectures.
Positioned as a neutral convener, Singapore is leveraging its unique East-West ties to foster dialogue where superpowers often default to rivalry. As the AI arms race intensifies, this consensus signals a fragile but vital step toward aligning global efforts on howānot just how fastāAI should evolve. ā Read the full article here.
š¾ FORWARD FUTURE ORIGINAL
The Second Half of 2025: What Can We Expect in the Area of AI?

It's December 2025: between a four-column PDF study, a database with 800,000 documents and an ongoing video conference, an invisible AI orchestrates the workflow, answers queries in real time, writes the final report and simultaneously reserves the train ride to the next appointment. What is still considered a tech-savvy demonstration today could become everyday office life within a few months.
The dynamics of the first half of 2025 certainly point in this direction: OpenAI has presented GPT-4.1, a language model that loads entire specialist libraries into a single session thanks to a context length of one million tokens, while Google is already looking to exceed the two million mark with Gemini 2.5. ā Read the full article here.
š BENCHMARKS
Why AIās Top Benchmarks Are Breakingāand How Social Science Could Fix Them

SWE-Bench, a go-to benchmark for coding AI, is facing backlash as developers optimize models for test performance rather than real-world problem solvingāa symptom of a larger āevaluation crisisā in AI. Critics argue that many industry-standard benchmarks lack validity, meaning they donāt reliably measure what they claim to, and can be gamed through shortcuts and selective reporting.
In response, researchers are turning to social science methodologies, emphasizing rigorous definitions and task-specific evaluations over vague measures of āgeneral intelligence.ā As AI capabilities grow faster than our ability to measure them, the tools meant to guide progress may be leading it astray. ā Read the full article here.
š¬ RESEARCH
AI Joins the Pack: Yellowstone Wolf Conservation Gets a High-Tech Howl Boost

A new partnership between The Colossal Foundation, Yellowstone Forever, and the Yellowstone Wolf Project is using AI and acoustic monitoring to decode wolf howls and improve conservation. By deploying 25 AI-enabled camera units near denning sites, researchers can now track not only the sounds wolves make but also the behaviors tied to those howls, providing rich ecological context in real time.
The machine-learning toolāalready 92% accurateāclassifies howl patterns to estimate pack size and identity, offering a less invasive, more scalable alternative to traditional tracking. Itās a leap forward in understanding wolves not just as symbols, but as vital ecological linchpins. ā Read the full paper here.
š°ļø NEWS
What Else is Happening

𦾠Fastino Shrinks AI Training: Startup raises $17.5M to train task-specific models on cheap gaming GPUsāfaster, cheaper, and laser-focused.
š Instacart CEO Joins OpenAI: Fidji Simo will lead Applications at OpenAI, stepping down as Instacart CEO but staying on as board chair.
š Claude Gets Web Search Powers: Anthropicās new API lets Claude browse the web in real time, signaling a serious challenge to Googleās search dominance.
š¶ Baidu Patents Pet Translator: The Chinese tech giant is developing AI to decode animal emotions into words, aiming for real cross-species communication.
š”ļø Google Adds AI Scam Shields: Chrome now uses Gemini Nano to detect scams in real time, blocking shady sites and scammy notifications before they reach users.
š§° TOOLBOX
Instant Websites, Interactive Stories, and Smarter Video Growth
šØāš» Durable: Instantly build pro websites with AIāno code needed. Includes SEO, marketing tools, and automated content generation.
š® AI Dungeon: Dive into AI-powered adventures where your choices shape dynamic, ever-evolving storiesāideal for writers and gamers alike.
ā¶ļø Agent Gold: Boost your YouTube channel with weekly AI-curated video ideas, SEO tools, A/B testing, and smart content optimization.
š¤ FRIDAY FACTS
An AI Image of the Pope in a Puffer Jacket Sparked Global Misinformation Panic

In 2023, an image of Pope Francis wearing a stylish white Balenciaga-style puffer coat went viralāonly it wasnāt real. It was created using the AI tool Midjourney, and it fooled millions, from Twitter users to fashion commentators. The image was so convincing that it prompted urgent calls for watermarking AI-generated content, sparking new policy discussions at both government and platform levels.
Now, in 2025, this moment is viewed as one of the first mass-scale "AI optical illusions," marking a turning point in public awareness of synthetic media. The European Union and several U.S. states have since enacted laws requiring disclosure labels on AI-generated images used in political or commercial contexts.
Why is it memorable? Because for many, it was the first time they realized: not everything you see onlineāeven a fashionable popeāis real.
Thatās a Wrap!
š°ļø Want more Forward Future? Follow us on X for quick updates, subscribe to our YouTube for deep dives, or add us to your RSS feed for seamless reading.
Thanks for reading todayās newsletterāsee you next time!
The Forward Future Team
š§āš š§āš š§āš š§āš
Reply