35,595 results · Most relevant
You cannot reserve token pools by a fixed ratio if you must serve many different workload context ratios with one engine or limited resources — you must have a unified way to manage all the low-level memory pools.
Hari Kumar recommends programming the internal platform specifically for each individual analyst and their model pattern — not one-size-fits-all — so it learns continuously from analysts' habits, and pulling data subscriptions with MCPs and APIs into that platform to increase their value.
Hari Kumar recommends measuring AI ROI by asking how long it would have taken to source information manually versus doing it in a more agentic manner, and by asking whether you would have reached the same conclusion with or without the AI tools.
Dane Knecht recommends auditing software and cutting code footprint so there are fewer places for vulnerabilities to exist, calling it the easiest first step. He notes AI-generated code tends to be verbose and complexity-creating, which currently works against secure software, and advises pushing back on over-engineered AI output — e.g., asking for the same result with roughly one-tenth the complexity.
creator advice, endorsed with high confidence
Jarmak advises choosing evaluation tasks that mirror the real workload and grouping them into task families so tests stay representative.
Jarmak recommends gathering as much observability data as possible and retaining all traces so evaluation and debugging have full context.
State clearly on the landing page what the product is and what action you want people to take, because visitors should not have to scroll halfway down the page to understand it. Focus landing page copy on the user's problem and how the product solves it, not on who the company targets. Talk about your product, who should use it, why it's valuable, and what problem it solves — not pitch investors. Use specific examples in your copy so users understand what the product actually does.
Spend time on the extra 20% rather than just shipping the vibe-coded output without taking it to the next level. If you don't balance speed with polish, you might have tons of users on your site who leave because they don't understand it and their workflow is blocked. You have to have faith that taking your time to think through design is worth it — the speaker's company has tremendous faith that it is worth it.
conditional — 'might have' users leave
The host recommends using specific language on landing pages about who should use the product and why, because vague generalized copy confuses users. Writing vague copy like 'institutional intelligence and decision infrastructure' leaves users asking 'is this something I need?' and unsure what the product actually does. He also recommends including specific examples so users can understand what the product does, and checking while scrolling whether they give specific examples.
The host lays out an upper body progression: start with standard push-ups executed as well as possible, progress to push-up jumps (a small hand jump, no clap), recruiting on the catch and applying force quickly, add a double bounce to trigger the stretch shortening cycle for more force, do explosive push-ups to a target (24 inch boxes recommended; the speaker uses 12 inch) because intent is higher and coordination happens at higher speed, vary target height, and finish with X push-ups (rebound push-ups). Push-ups to boxes are recommended for high speed force, reaction, and coordination.
The host lays out a lower body progression: start with split squats (three reps per side, feeling the glute and hip), progress to walking lunges (adding steps, even backwards), then to single leg squats with the foot behind the glute, 60-70% of load on the front leg, five to seven reps.
The host lays out a full-body speed progression: start with easy skips (controlling the abs and trunk; skips are a foundational movement to do all the time), then single-leg hops (five per side, working the grounding position and reaction, hip lock position, double bounce on landing, trunk tension or you fall sideways, trunk control to coordinate the upper body into a double-leg landing), then bilateral broad jumps for distance (land almost flat-footed, sticking the landing; serious force produced quickly), and finally broad-into-vert jumps (reload the upper body in the air to change center of mass and direction upward). Variations can be paired and played with.
Huberman stresses consulting a trusted board-certified physician before adding or subtracting any behavioral, nutritional, supplementation-based, or prescription drug-based protocols, and says any attempt to reduce or eliminate prescription drugs must be done in discussion with physicians.
Consequential creator advice requires independent verification before Today can promote it.
Tanmay Sah recommends designing a materiality scale to identify which AI use cases are riskier than others, prioritizing use cases by materiality because materiality rank-orders with harm, and allocating more resources to higher-materiality use cases.
guest's recommendation, attributed to him
Brockman advises assessing whether a field is tractable for AI by asking whether you can generate lots of training data in it: math and theoretical physics have abundant data, while biology is data-bottlenecked. He also suggests checking whether other fields share the same enabling properties, since capability transfers where underlying conditions like data availability match.
Brockman argues that navigating the AI cybersecurity moment requires a coordinated response, and that defenders need to internalize how the physics of cybersecurity are changing — recognizing there is a window that will not last forever. He believes defenders can outstrip attackers in the new AI security world, but it will take a lot of change: a shift in how every organization operates and how we collectively think about critical infrastructure and resourcing those that have not had those resources to date. He also notes organizational security now depends on both one-time fundamentals and ongoing cyber capability used for defense.
The tool lever takes up all the air in the room and on Twitter because it is exciting new tech, but leaders have little control over it — Arcolano doesn't work at a foundation lab and can't make the tools better. As costs tighten, leaders have less freedom to buy the tools they want. The things leaders can control are on the other side of the equation: the people and the organization. Controlling company/team/culture factors is the leadership job, even if the CFO caps token spend. And unlike golf, you can change the course: leaders can change their engineering environment.
Build a pre-call crib sheet: block an hour at the start of your day (if back-to-back) and compile who you're meeting with plus company research — press releases, stock news, anything notable. Pre-call research only takes 5-10 minutes — you don't need a monstrous amount of time to learn what the company does and what's notable about it.
Send a handwritten thank you note when you lose a deal — it demonstrates character and positions you as the first call when the chosen vendor fails or the buyer changes jobs. Don't push back with 'what do I have to be thankful for?' — you're playing the long game, and the winning competitor probably didn't send one, so the note sets you apart.
The source recommends that instead of going for someone you think could be good, you find creators that already know how to make content — they know the basics of what they're doing — and then sell them on why your brand is the best opportunity for them.
endorsed
Choose sources → to get a library of your own and a daily brief.