- Bay Area Times
- Posts
- iOS 26, macOS Tahoe among others to debut at Apple’s WWDC 2025 at 1 PM ET today
iOS 26, macOS Tahoe among others to debut at Apple’s WWDC 2025 at 1 PM ET today
Bay Area Times
June 09, 2025
Brought to you by:
Top stories today:
- iOS 26, macOS Tahoe among others to debut at WWDC 2025 today
- Apple researchers find “reasoning” models collapse
- Microsoft, Asus unveil ROG Xbox Ally, Xbox Ally X gaming handhelds
- Meta said to make massive investment in Scale AI, possibly $10B+
- Elon-Trump feud complicates xAI $5B debt raising with Morgan Stanley
0. Data and calendar
All values as of 6 AM ET / 3 AM PT, other than S&P500 and NASDAQ close (4 PM ET / 1 PM PT).
All times are ET.
Listen to our AI-generated podcast summarizing today’s newsletter (beware of hallucinations):
1. iOS 26, macOS Tahoe among others to debut at Apple’s WWDC 2025 at 1 PM ET today
Liquid Glass interface is rumored to overhaul Apple’s software experience.
New software nomenclature (e.g. iOS 26) this year.
Apple’s LLM access to 3rd-party developers may debut.
Apple Intelligence AI strategy may be discussed at the keynote.
Major updates to Apple apps are incoming, including games as a pre-installed app.
Plaid’s biggest launch of the year is here—with 20+ product updates across fraud, credit, payments, and more going live June 12.
Join thousands of fintech and enterprise leaders for an early look at:
Our most powerful fraud platform yet
New cash flow insights for smarter lending
AI-powered tools for smarter workflows
If you care about stopping fraud sooner, approving credit smarter, or scaling faster—this is the one to attend. It’s free and fully virtual.
*Sponsored.
3. Apple researchers find “reasoning” models collapse beyond complexity thresholds
“Reasoning” models fail on harder problems, even when they have tokens left, the researchers noted.
General LLMs outperform reasoning models on low-complexity problems, the study revealed.
However, “reasoning” models outperform general LLMs when problem complexity is medium.
The researchers considered 4 puzzles: Tower of Hanoi, Checker Jumping, River Crossing, and Blocks World.
OpenAI’s o3-mini (high and medium), Anthropic’s Claude 3.7 Sonnet (non-thinking and thinking) and DeepSeek R1 and V3 were the models tested.
Some believe that while the Apple paper is useful, it doesn’t support the idea that LLMs are “hitting a wall”
Other commentators said:
Showed LLMs fall short of AGI and can’t replace well-defined conventional algorithms: scholar and GenAI critic Gary Marcus.
Highlighted that the deeper problem isn’t LLM performance, but the persistent hubris that has shaped AI since its inception: Steven Sinofsky.
Overtrusting “reasoning” AI models could be risky, even though they often outsmart mathematicians.
