Explore

Self-assessing movement: joint-by-joint video review for symmetry and stability

Dr. Galpin's locally coherent method for movement self-assessment: (1) ideally see a highly qualified physical therapist or movement specialist; if self-assessing, (2) pick a representative movement for you (bench press if you bench, pull-up if you do pull-ups), (3) record frontal and side views, three to ten slow, uncontrolled reps per angle, (4) review joint by joint — shoulder, elbow, low back, hip, knee, ankle — looking for symmetry (front-to-back, left-to-right, one limb ahead or fidgeting) and stability (e.g., pausing three, five, or ten seconds at the bottom with complete control; shaking knees indicate instability), (5) score 0, 1, or 3: 0 = don't do the exercise due to high acute injury risk, 1 = minor flaw, proceed cautiously with load and volume, 3 = go ahead on a reasonable —

Consequential creator advice requires independent verification before Today can promote it.

Method, not a verified protocol.

Andrew Huberman · Essentials: How to Assess & Improve All Aspects of Your Fitness | Dr. Andy Galpin

Don't just stop GLP-1s: transition to microdosing or a bridging plan to keep weight off

Dr. Chris Thompson says stopping GLP-1s causes problems — these drugs are not meant to be stopped. For maintenance, rather than stopping, patients should transition to microdosing to find a spot where they keep the weight off, feel good, and take less medication; if coming off a GLP-1, you need either to microdose it or have a bridging plan to a procedure or something else that will keep the weight off.

Consequential creator advice requires independent verification before Today can promote it.

Andrew Huberman · Best Tools for Gut Health & Weight Loss | Dr. Chris Thompson

Ready-jump command structure for teaching weightlifting

When teaching weightlifting to a complete beginner, tell them to throw out everything they've ever learned in sport. Use a 'ready, jump' command structure: when I say ready, you get ready; when I say jump, you jump. The purpose is to manipulate space and time so the learner cannot think. In the ready-jump method, the learner lifts on the coach's time, not their own time, and has no space or time to analyze a failed rep because the next command is coming immediately. The most critical thing when teaching someone to weightlift is to take their brain out of it.

Method, not a verified protocol.

Garage Strength · Why Most People Train Explosiveness Wrong w/ Exercise Scientist Dr. Dan Cleather

Establishing foot position and the catch through natural jumping

Dan establishes foot position at the very start of teaching the lift. He starts by having the athlete simply jump on command, then has them take a step in and jump, and non-judgmentally observes where their feet land to discover their natural jumping position, which becomes the starting foot position. He then places the bar on the back of the neck and has the athlete jump and catch it in a quarter squat to establish the receiving position. In a group setting, athletes hold the quarter-squat catch until the coach has checked each one. He runs a block of 'ready, jump' reps from the back of the neck (like a push jerk) so the athlete owns the catch position before pulling from the floor. He ignores cosmetic 'ugly' technique details as long as the lifter is safe and is solving the target

Method, not a verified protocol.

Garage Strength · Why Most People Train Explosiveness Wrong w/ Exercise Scientist Dr. Dan Cleather

Drilling the first pull: deadlift to above the knee, then add the jump

To teach the first pull, Dan has athletes practice deadlifting the barbell to just above the knee and drills that until it is grooved, then asks them to do that and jump. He alternates one rep of pulling the bar to above the knee with one rep of pulling to above the knee and jumping to reinforce the pattern. The first pull is cued with three simultaneous checkpoints: knees move back, hands follow the knees backwards, and the back comes up at the same angle. He also cues 'big chest' repeatedly during the first pull to keep the lifter safe, and uses 'knees back' and 'jump' as the primary floor-pull instructions.

Method, not a verified protocol.

Garage Strength · Why Most People Train Explosiveness Wrong w/ Exercise Scientist Dr. Dan Cleather

Teach power versions first, snatch before clean, and progress from high hang

Dan would always teach the power versions of the lifts first, before the full squat versions, and drills the catch position before introducing jumping with the bar. Most of the time he teaches people to snatch first because the power-snatch catch is easier than receiving a bar at the neck as in the clean. Rushing to lift from the floor is unnecessary; spending weeks on high-hang snatching is valid and valuable. He spends only about 10 minutes on hang snatches at the start of a session rather than dedicating the whole session to learning the lift, and has taught a group of basketball coaches the snatch in 15-20 minutes using repeated 'ready, jump' reps progressing down to just above the knee.

Method, not a verified protocol.

Garage Strength · Why Most People Train Explosiveness Wrong w/ Exercise Scientist Dr. Dan Cleather

Breaking technological addiction with device-free windows

Brooks advises breaking the cycle of technological addiction by keeping the first hour of the day, the last hour of the day, and mealtimes device-free. He explains that the first hour matters because neurocognitive programming happens in your first hour awake, and the last hour matters because relationship programming and sleep architecture are wired in the last hour of your day.

endorsed by Brooks

Method, not a verified protocol.

Andrew Huberman · The Art of True Happiness | Dr. Arthur Brooks

Receiving and observing gratitude is the most potent way to raise serotonin

Behavioral, nutritional, and supplementation tools can tap into the serotonin system to positively impact feelings of wellbeing, though with less potency than prescription drugs. Physical contact such as holding hands, hugs, and cuddling can increase serotonin transmission and makes people feel good. Receiving — not giving — gratitude has the most potent effects on increasing serotonin and serotonergic brain circuit activity, and observing others giving and receiving gratitude is also immensely powerful. Receiving and observing gratitude is the most potent way to increase serotonin in the brain and body.

Consequential creator advice requires independent verification before Today can promote it.

Andrew Huberman · Control Your Brain Chemistry for Focus, Motivation & Well-Being | Huberman Lab Essentials

Delegate diagnosis so the founding surgeon focuses only on diagnosis

Hormozi advises Hussey to delegate diagnosis work to another doctor so his dad can focus only on diagnosis, saying that change alone can get the business to roughly $12 million.

Consequential creator advice requires independent verification before Today can promote it.

Alex Hormozi · This Video Should Be Required Viewing for Business Owners

Establishing identity and delegated scopes

First establish who the user is: an authorization server verifies identities and issues tokens (Google, Okta, and Auth0 provide authorization servers). If you want an identity chain in the agent execution path, the user has to log in first. The authorization server prompts the user for consent to delegate access to resources with a subset of their permissions (scopes) — the first narrowing of access — and only some of the user's total scopes should be delegated to the agent. The server issues tokens identifying the user and their level of access, so you know who the user is and what they can access — better than the demo.

Method, not a verified protocol.

AI Engineer · Your Agent Is a Lethal Trifecta (And What to Do About It) — Kim Maida, Keycard

Workflow classification: what the LLM is doing right now, from timing

The workflow classifier answers 'what is the LLM doing right now' on the same channel, distinguishing four workflow types: prose, code, JSON, and retrieval. It makes a guess every half second based on the last five seconds of timing; out of 40 half-second guesses 31 were right, and four full responses were correctly identified by majority vote. If the attacker already knows the model from the earlier slides and spectrally recorded which workload it happens to be on, it can also tell what kind of text is coming out.

Method, not a verified protocol.

AI Engineer · Your GPU Neighbor Can Hear Your Model — Branislav Brzak, Stealthium

Layer-count fingerprinting: fold-and-average once per model to count layers

The layer-count method folds and averages the tick signal once per model instead of once per token, producing a bit count that lands almost exactly on a whole number. The mean distance from an integer across 22 traces is just 0.028 — that's the tell that the count is a real layer count, not a hardware artifact. Each count works out to the model's layer count plus 1 to 3 additional peaks. If the layer count signal were just a fixed hardware artifact divided by different token rates, we'd expect fractions, not whole numbers — the whole-number result is the tell.

Method, not a verified protocol.

AI Engineer · Your GPU Neighbor Can Hear Your Model — Branislav Brzak, Stealthium

The layer-fit classifier: a best-fit depth search, not a trained ML model

The layer-fit classifier is not a trained machine learning classifier: there is no training set, no labels, no configuration, and it is under 30 lines of code. The method checks each candidate's depth from 16 to 48 layers, measures how strongly that depth's expected signature shows up in the real trace, and picks the best fit. For Qwen3 8B, 36 layers wins clearly by 4.0 times over the next candidate.

Method, not a verified protocol.

AI Engineer · Your GPU Neighbor Can Hear Your Model — Branislav Brzak, Stealthium

The attacker's probe primitive: time your own memory reads

The attacker's probe primitive consists of three steps: load from its own small buffer, time the load, and repeat every 5 microseconds. It only ever reads its own memory and never touches the victim's memory. With the naked eye, comparing a co-tenant idle versus serving an LLM, you wouldn't be able to discern the timing differences, but a median latency shift of about three microseconds is enough for the attacker.

Method, not a verified protocol.

AI Engineer · Your GPU Neighbor Can Hear Your Model — Branislav Brzak, Stealthium

The agent registration flow: identity assertion, AuthMD discovery, IDJAG credential, access token

The source describes the registration step as where the magic happens in the agent authentication flow: the agent mints an identity assertion and gives it to Cloudflare, which can then verify it — or not — and can choose to verify out of band; the auth is very flexible. The agent stack includes your LLM, maybe your memory, and different context layers, and connects to your backing identity service (the user identity). When the agent registers, it sends the discovery call to Cloudflare looking for the authMD service. The agent sends the IDJAG (identity JSON access grant) — essentially a signed credential that represents the user — and receives back the access token (actual API token), and then registration is done. After registration, you can make normal API requests, call MCP, and everying

Method, not a verified protocol.

AI Engineer · https://www.youtube.com/watch?v=nSkmXakJ6eo

Build a context layer for code with the extract-persist-query recipe

To build a context layer for code, follow the same extract-persist-query recipe as for data, scaled down to a runnable size: extract information about every symbol — every type, every trait, every function — with its real identity and location, and represent relationships between symbols as edges in a graph that can be traversed; compute them once and query them many times.

Method, not a verified protocol.

AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef

Two-phase athlete assessment: strength first, then technique

Assess foundational strength first as a simple ratio to bodyweight, taking into account the athlete's current level and training status, and decide whether strength needs to be a priority. If strength numbers are already adequate for the goal, move to the second phase: the technical aspects of the movement or sprint — altering patterns, bringing in accessories or derivatives, and changing how the athlete runs or moves.

Method, not a verified protocol.

Garage Strength · How To Get Strong and Still Be Athletic w/ Sika Strength

Air bike conditioning protocols: intervals, steady state, and testing

For improving conditioning on the air bike: do 10 sets of 10 calories or one set of 100 calories once or twice a week, intensely; the same workout week to week with genuine effort produces good progress without elaborate programming. For longer work, do 60-90 minutes of long slow distance keeping heart rate at 70%, or interval work of six by six minutes at 85-90% heart rate without going above that ceiling, and occasionally push past threshold into the zone where you feel like you are about to die. You can get an excellent conditioning session in under 10 minutes.

Endorsed protocols from the guest

Method, not a verified protocol.

Garage Strength · How To Get Strong and Still Be Athletic w/ Sika Strength

Use an agent firewall to block commands that haven't been blessed

Patterson recommends an agent firewall to ensure the agent cannot run commands you haven't already blessed: it inspects process logs and denies dangerous commands like 'gh repo delete' or dropping a table.

endorsed

Method, not a verified protocol.

AI Engineer · The Lethal Trifecta Is Already on Your Laptops — Michael Patterson, Coder

Synthesize queries by passing schema, current time, and user info to the LLM

Query synthesis takes the dynamically resolved schema, the current date and time, and information about the asking user, then asks the LLM to synthesize a query. Including user information in the query synthesis context lets the engine resolve questions like 'my PRs' instead of only 'Peter's PRs'. Query synthesis is roughly the easiest part of building the structured retrieval pipeline.

endorsed

Method, not a verified protocol.

AI Engineer · Beyond RAG: A Relational Context Engine That Cuts Token Burn — Unblocked