37,665 results · Most relevant
Setup guidance for the Jefferson deadlift: place the lead leg forward with about a 90-degree angle on the other side, position the bar slightly closer to the front of the lead leg, stay relatively close to your conventional deadlift stance width (adjusting for hip mobility), and perform the lift on both sides.
A locally coherent method for the hack lift: get set by sticking your chest out, sticking your knees out, and getting a little wave rolling so the tension goes right to the quad — the movement is almost like doing a little wave forward, which loads the quads heavily. To avoid it hitting your calves, drive the knee forward, which loads the quad even more. You can cheat the lift by putting a wooden board or plates under your heels, and you can raise the height of the plates if you want.
A locally coherent method for the clean and press: set the feet about a foot width apart, take the grip with the crook of the elbow right outside the knee to establish the clean position, power clean the bar and widen the feet just slightly to get into the receiving position, keep the elbows and lats mobile up top in the rack, hold the bar in the rack with an open palm dip, then press — essentially like a push press to a point — without re-bending the knees or hips, driving straight back overhead.
Quigley advises against training drug-binding models on public chemistry datasets because the data reflects the sociology of what got made and tested rather than ground truth chemistry — 'when you train models on the exhaust, you might be in trouble.' As a related diagnostic, he suggests replacing chemical structure representations with a vector of creator-identity probabilities to test whether a model is learning chemistry or just creator sociology.
Consequential creator advice requires independent verification before Today can promote it.
Leash Bio physically mutates the DNA construct of a wild-type protein already screened, changing one or more amino acids in the functional binding region (e.g., inserting a bulky residue to fill the pocket or swapping a side chain), then re-runs the DEL screen. Typically three to five amino acids drive binding in a pocket; they break them independently and redo screens each time to learn which part of the small molecule interacts with the changed residue.
Hari advises defining a 'novelty budget': if you can get at least one unit of true novelty — like PageRank was for Google — you can cross the median for a good amount of time. Before the build starts, name the one thing that is novel — if you don't have one, take the median because solutions for your problem already exist. He says 'you tell me a problem for which there are no solutions, I'll go and find it for you' — every problem already has solutions. Put the novelty plan as part of your roadmap and project planning, and review it the way you review spending — like a project. If you cannot name the novelty in your project, concede that it's a learning exercise — that's fine. After two years of burning several million tokens, everything else is free now — the highest ROI is the novelty
endorsed
Each chunk carries metadata including rulebook name and page (for citations and PDF links), a chunk index (for reassembly), business-logic fields, and an LLM-generated rulebook summary so the agent knows the document context.
Text is extracted per page, split into paragraphs, packed into ~2,000-character chunks, and section headers are detected with regex so every chunk carries its section path.
Jemima Abu advises enforcing a mandatory accessibility skill that AI agents use anytime you run anything related to generating UI or writing code that requires accessibility.
The first decision most teams get wrong is data sampling: hold a soft target fraud rate of 0.5% for card-related merchant transactions (about 10x the 0.05% natural rate) and 6% for online transactions (about 10x the 0.6% natural rate). Both targets are far gentler than the aggressive over- or under-sampling a tabular pipeline would use.
Specific targets attributed to MD Sharique.
It would be an extremely positive development if the labs tried their best models on basic cryptographic assumptions and the models failed to break them, and the labs disclosed that failure. Labs could spend $20-50 million running their most cyber-capable unrestricted models to try to break cryptographic systems. Another thing the labs could try is using these models to give much stronger evidence that cryptographic problems are hard. These are pure math questions that frontier labs could be turning their models on.
The $20-50 million figure is a hypothetical estimate.
Dr. Thompson recommends that people he is concerned about regarding muscle loss get a DEXA scan beforehand to confirm adequate muscle mass; if the scan shows inadequate muscle mass, he says you really have to think twice about whether to do the GLP-1 or try another avenue to lose the weight.
None stated.
Dr. Chris Thompson recommends getting a continuous glucose monitor for a month or two — not all the time — to learn what spikes your glucose, then adjusting your eating accordingly. A CGM can catch excess glucose early and reveal whether you have particularly large glucose spikes to certain foods. If your glucose shoots up to 200 with certain meals, you know you're sensitive to that food. If a food spikes your glucose, you should change how you're eating it — such as eating it after having something fatty — or maybe avoid it. Glucose spikes are part of a sequence that will lead to problems, so they are worth acting on.
Dr. Thompson recommends getting screened for the structural eye condition (optic nerve head ischemia) before taking GLP-1s, while emphasizing it is not true that GLP-1s are making people go blind all over the place.
None stated.
Dr. Chris Thompson advises starting to look early for metabolic problems and not relying on the traditional tests — look at other things. The next phase in the metabolic dysregulation sequence is high fasting insulin levels, which you can check with an inexpensive fasting insulin test. He notes that waiting until you have diabetes and your A1C goes up is the wrong approach, and that there are many things we should be doing before the standard tests of fasting glucose and hemoglobin A1C.
The creator recommends overlaying power clean VBT data with the athlete's speed component, single leg squats, and back squat numbers for deeper programming.
A step-by-step teaching progression for the power clean: start with the bar at the thigh, then move to just below the knee, then to the middle of the shin, performed with an empty bar before adding weight. Each position has specific cues: from the thigh, bend with butt back and knees slightly bent, then pull to the rack position without dropping under; from just below the knee, think 'knees come back, make contact, catch' performed slowly; stay flat-footed as long as possible during the pull. The progression is recommended as a daily warm-up to focus on mobility.
The creator describes an in-season approach for George: put on 110 to 120 kilos (about 255 pounds), hit four doubles, and maintain strength through the season based on speeds and percentages, with around 90% of his speed work meaning he maintains strength.
Create tiered service levels with different mover skill levels (e.g., level 1, 2, 3 movers, where a level 3 has done around a thousand sessions). A VIP package includes master movers, packing/unpacking, and all materials; a standard package includes packing but uses non-master movers.
asserted
A complete 15-minute mobility session built around the overhead squat: 10 minutes of warmup followed by 5 minutes of progression work, starting with a dowel rod and building toward loaded overhead positions.
Choose sources → to get a library of your own and a daily brief.