inshard

Nice page design and even nicer photography.

The UI is better - they box the specific types of actions the orchestrator agent takes with a clear categorization. The standard quality of life shortcuts like type a number to respond to an MCQ are present here as well. They use specialized sub agents such as one with big context window to find context in the codebase. The quotas appear to be much more generous vs CC. The agent memory management between compacting cycles seems to have a few tricks CC is missing. Also, with 3.0 Flash, it feels faster with the same level of agency and intelligence. It has a feature to focus into an interactive shell where bash commands are being executed by the orchestrator agent. Doesn't feel like Google is trying to push you to buy more credits or is relying on this product for its financial survival - I suspect CC has some dark patterns around this where the agents runs cycles of token in circles with minimal progress on bugs before you have to top up your wallet. Early days still.

Tested it on Gemini CLI and the experience as good if not better than Claude Code. Gemini CLI has come a long way and is arguably likely to surpass Claude Code at this rate of progress.

This segment about the mechanism is simple and very profound. I wonder if any cancer researchers here could comment on its universality across various types of cancers:

"Tumor-Specific Accumulation Mechanism

E. americana selectively accumulates in tumor tissues with zero colonization in normal organs. This remarkable tumor specificity arises from multiple synergistic mechanisms:

Hypoxic Environment: The characteristic hypoxia of tumor tissues promotes anaerobic bacterial proliferation

Immunosuppressive Environment: CD47 protein expressed by cancer cells creates local immunosuppression, forming a permissive niche for bacterial survival

Abnormal Vascular Structure: Tumor vessels are leaky, facilitating bacterial extravasation

Metabolic Abnormalities: Tumor-specific metabolites support selective bacterial growth

Excellent Safety Profile

Comprehensive safety evaluation revealed that E. americana demonstrates:

Rapid blood clearance (half-life ~1.2 hours, completely undetectable at 24 hours)

Zero bacterial colonization in normal organs including liver, spleen, lung, kidney, and heart

Only transient mild inflammatory responses, normalizing within 72 hours

No chronic toxicity during 60-day extended observation"

This is my theory as well. A google search for the late prof's name returns a .ir website at the top of the result for some reason. It's a tragic loss for the world and his loved ones as are the victims of the brown incident.

Strange that a 9th October article shows up at the top of the feed given the events of the day in Sydney.

Everything is relative. Even the perception of effort, from the calories burned at work daily to sustain a livelihood, is subjective. What truly matters is the amount of effort required by your peers to achieve similar financial stability. We tolerate the work as long as everyone else is equally willing to do it.

Seattle hits a doom (helped by the gloom) loop every winter. This too shall pass.

This is the worst these things are going to be.

Thorium is abundant in Sri Lanka’s mineral sands. Mined with dredgers at shallow depths 10-100m off the western coast.

I was surprised when Microsoft hired him. He always seemed to be the cofounder to Demis Hassabis who took on a philosophical tangent rather than the hard engineering needed to build transformative technologies and user experiences. I feel Microsoft lost a gem when Panos Panay left them. He really did some great work on the Surface product line. Surface Studio in particular.

Yep, this reinforces my point. Range Rovers try to maximize scope, often at the expense of scalability and reliability. Same with Bentleys, the Concorde passenger jet and the Harrier fighter jet.

Could it be guessing via orders of magnitude? Like 145 lb * 1/4 is high confidence not the answer and 1 lb 45 oz is non standard notation as 1lb = 16oz - so it's most likely 14 lb 5 oz

US has a culture of scaling. Having lived in London and many parts of the US, I could observe that Europe has economies of scope. And the US has economies of scale. A good example would be the Ford F150 and the Range Rover. One is highly reliable, built to capture the core value of a Utility vehicle, as cheaply, and as numerously as possible. The other tries to be the best possible version of an SUV. It’s not that the F150 is playing to the mass market. It’s just the maximum features, that doesn’t compromise scale. Many other examples like Southwest Airlines, McDonalds, manhattan’s skyline, Arm and Hammer, US highways, US Domestic Airports, Nike Vemeros, CVS, US Healthcare Triage and Emergency Services. It’s just less fancy, more accessible wealth creation.

AKA the OG Pixel Watch 4

Do we need to distinguish between Atlas, the control theory based humanoid tech stack vs say Optimus that may run on an end-2-end visual reasoning world model? I feel these are two different paradigms. Same for military drones. The next gen prototypes may have launch-and-forget levels of autonomy, however ethically questionable.

If we are thinking about world models and embodied AI as the next big wave, I feel we need to factor the size of this market across military and civilian applications. When I hear Tesla AI4-6 plans for autonomous driving and Optimus, Apple’s home robot, and many such initiatives, I’m also thinking of an autonomous military drone arms race and autonomous space exploration and moon colonization. Edge compute will be a key innovation and I feel Apple is gonna be best placed to exploit their power efficient M-series architectures.

Whenever in doubt about your product’s acceptance, just jack up the price and keep a straight face.

Yep. Essentially take risks, expand your world model, but above all, don’t die. There’s a tension there - like “what happens if I poke the bear” vs “this might get me killed.”

Also good context here is Friston’s Free Energy Principle: A unified theory suggesting that all living systems, from simple organisms to the brain, must minimize "surprise" to maintain their form and survive. To do this, systems act to minimize a mathematical quantity called variational free energy, which is an upper bound on surprise. This involves constantly making predictions about the world, updating internal models based on sensory data, and taking actions that reduce the difference between predictions and reality, effectively minimizing prediction errors.

Key distinction: Constant and continuous updating. I.e. feedback loops with observation, prediction, action (agency), and once more, observation.

It should have survival and preservation as a fundamental architectural feature.

The author misses the science of emergence. Reductionist views can’t fully explain macro-level capabilities that arise in these systems. Something emerges at higher scales from the possibility space as model sizes grow; they stop being mere “stochastic parrots” or black boxes running simple regressions.

The weights develop their own inherent logic based on how they relate to each other, analogous to how brain waves encode memory at a level higher than individual neuron networks.

Ultimately, the value of AI lies in the imagination of its wielder. The Unknown Unknowns framework is a useful tool for navigating AI effectively (it powerful to help elaborate on Known Unknowns and identify Unknown Unknowns), along with a healthy dose of critical thinking and understanding how reinforcement learning and RLHF work post-pretraining.

The article ignores agriculture and animal husbandry as the critical inflection point for property rights and mass conflict. You could argue proto-versions of these dynamics existed even in abundant hunting grounds—see Native American tribal warfare over territory. More importantly, it misses technology as a third layer. As technology deflates the base layers of Maslow’s hierarchy, inequality matters less in absolute terms. The main constraint now is housing, but even that could be solved through vertical development and better long-term urban planning. We might actually be at escape velocity to break the historical cycles described here.

I’ve occasionally seen and understood physical shapes and concepts while dreaming, which I was unable to describe or articulate in the real world once awake. It’s as if I recall how it made me feel, but I can’t describe what it was. Like the memory of a face you saw just once.

Same sentiment. It’s still touch through a slightly less solid medium.

This misses the fundamental information problem. Your recommendation algorithm is centralized—it only knows what its signals can measure. Ads create a decentralized market mechanism where businesses themselves can signal “your algorithm is underweighting me.”

Consider the failure modes of pure algorithmic ranking:

Cold start problem: A phenomenal new restaurant opens. It has no ratings, no historical visit data, no repeat customer signals. Your algorithm buries it. How does it escape this trap? Organic discovery is glacial—it might take months to accumulate enough signals while the business burns cash.

Structural bias: Your algorithm might systematically underweight certain business types. Maybe sit-down restaurants generate longer “time spent” signals than excellent quick-service spots. Maybe your visit detection misses certain building types. The algorithm doesn’t know it’s biased.

Local knowledge asymmetry: The business owner knows their value proposition intimately—they know their recent quality improvements, their new chef, their differentiation. The algorithm is looking backwards at historical data.

Network effects lock-in: Once a place is highly ranked, it gets more visits, more ratings, reinforcing its position. Even if quality declines, the algorithm is slow to react.

Quality-weighted ads let businesses with superior local information challenge the algorithmic ranking. If you’re genuinely better than your algorithmic position suggests, you can bid to prove it. The quality weighting means you only profit if you’re right about your own quality—it’s costly signaling backed by conversion economics. This is “outside-in” because you’re not trying to perfect a centralized algorithm. You’re creating a market mechanism where distributed information surfaces through economic incentives. The businesses that are most undervalued by the algorithm have the strongest incentive to correct it.

Pure algorithmic ranking is central planning. Quality-weighted ads are a market.

The value of AI is in the imagination of its wielder. The Unknown Unknowns framework is a useful tool in how to navigate AI along with a healthy dose of critical thinking and understanding how reinforcement learning and RLHF work, post pre-training.

Local businesses with better quality usually have better ratings in maps and better economics—higher margins, repeat customers, lower acquisition costs. And since only nearby places can compete, you get real competition on merit instead of a race to the bottom with faceless actors. Good ads solve a real problem: helping people discover great spots in unfamiliar cities.

Jobs saw something with iAd.

The problem is simple auction mechanics favor whoever has the deepest pockets. A mediocre chain with fat margins outbids an amazing local place, even if the local spot delivers way more value. You’re optimizing for who can pay, not who’s actually good.

To fix this, you weight bids by quality signals like ratings, time spent and repeat visits.

Now ads amplify what’s already great instead of just selling visibility.

Users get better recommendations, good businesses win, and Apple builds trust. That’s how you turn ads from a tax on attention into actual product value—and an improved user experience.

Unions are like winning the battle to lose the war, in free markets. A non unionized or automated competitor eventually kills the unionized incumbent.

It is whataboutism if the reply was a reply to my post about the UN resolution denouncing the sexual slavery and exploitation of minors in the Quran. Which seems to be the case as the argument was specific to the UN resolution statement that I made in my comment. Not the OP’s argument.

Funding a seed stage VC with a record of picking winners like Sequoia is probably very easy even without the few sovereign wealth funds from the Middle East. Sequoia’s checks are mainly in the early rounds where the sizes are relatively smaller to series C+. The American university endowments would oversubscribe their rounds multiple times over.