Because a scaling law is only fit on the (relatively small, relatively cheap) models that we can afford to train, and the prediction is _extrapolated_ for a model orders of magnitude larger.

Those deals with Anthropic and Google have relatively favorable termination clauses for SpaceX, so if SpaceX's enterprise AI efforts take off and see high demand, it will theoretically be possible to reallocate compute from competitors directly to SpaceX and the Cursor team.

OpenAI选择砍掉视频应用，把算力集中到GPT-5.5的Agent架构和Codex代码工具上

Google fellow John Jumper, who won the Nobel for AlphaFold, is now working on AI coding, not on science-specific AI tools

Commoditizing complements doesn't always work because focus is scarce even for the largest, fastest growing businesses.

Commoditizing complements doesn't always work because focus is scarce even for the largest, fastest growing businesses.

Commoditizing complements doesn't always work because focus is scarce even for the largest, fastest growing businesses.

Structure: $500M OpenAI equity plus $4B from TPG, Bain, Advent, Brookfield, and Goanna form a $10B LLC.

At our request, the underwriters have reserved up to _______% of the shares of Class A common stock offered by this prospectus for sale at the initial public offering price through a directed share program to certain persons identified by our management and certain long-tenured employees, which may include parties with whom we have a business relationship and friends and family of management and such employees.

The real bottleneck in AI right now is not compute but rather data quality

The third allocation strategy is to allocate each function in a way that maximizes economicefficiency.

The second allocation strategy is assigning each function to the most capable agent, which canbe either a human or a machine.

Therefore, this automation strategy definesthe roles and responsibilities of users in terms of automation instead of the other way around.

First, byautomating everything that can be automated, the user is left with functions that, by definition, thedesigners find hard, expensive, or difficult to automate.

The first strategy is called maximum automation. Here, each task that can be automated is allo-cated to a machine.

The aim is to increase efficiency, reduce costs, or both.

Which system functions should beautomated, and to what extent should they be automated?

Task allocation is a central challenge in HCI and automation.

Independently of the chosen strategy, some tasks are to be done by the interactive system andsome by the user. The allocation of such tasks is called functional allocation.

In all cases, getting the right balance betweenautomation and human control is crucial.

In HCI, automation poses the foun-dational problem of task allocation: Who does which task or subtask—the system or the user?Moreover, how much should we automate?

approaches for translating each ESB to cities and businesses via the sequential steps of transcription, allocation, and adjustment.

Ryberg et al. (2018) explicitly looked into the choice of sharing principles (also referred to as allocation principles or distributive principles in other literature).

The policy provides a simple test of an RA’s competence: the ability to make a DOI Kernel Declaration, which requires that the RA has an internal system which can support the unambiguous allocation of a DOI name, and is fundamentally sound enough to support interoperability within the network.

Approximately sixty formal URN namespace identifiers have been registered.

La mise en place d’une telle « allocation d’autonomie » est défendue en France91 depuis la Charte de Grenoble de 1946 par des organisations comme l’Union nationale des étudiants de France, mais n’a jamais été réalisée. La mesure a failli être adoptée en 1951, alors que « la commission de l’Éducation nationale de l’Assemblée nationale avait ainsi adopté à l’unanimité le rapport Cayol en faveur d’une rémunération étudiante92 ». Le refus d’adopter la mesure en France s’expliquerait notamment par l’influence croissante de la théorie du capital humain néolibérale qui présente l’étudiant·e comme un investisseur en lui-même93. Le salariat étudiant était encore revendiqué en 2018 en France (ainsi qu’au Québec par les Comités unitaires sur le travail étudiant94) et l’idée a été notamment reprise par le candidat présidentiel Jean-Luc Mélenchon à la suite de l’immolation d’un étudiant pour cause de précarité (celui-ci avait écrit un texte revendiquant le salariat étudiant avant de poser son geste). Mélenchon proposait dans sa plateforme une allocation de 800 euros par mois95.

Or the PIs who enjoy and excel at raising funds can do so and even re-deploy it to the right scientists, akin to founders who become angel investors and venture capitalists.

Computer scientists make LDA seem complicated because they care about proving that their algorithms work.

Having made these points many times in the last few years, I've realized that the fundamental problem is in the mistaken belief that the type system has anything whatsoever to do with the storage allocation strategy. It is simply false that the choice of whether to use the stack or the heap has anything fundamentally to do with the type of the thing being stored. The truth is: the choice of allocation mechanism has to do only with the known required lifetime of the storage.

If we’re in that situation when new memory is allocated then the “high water mark” is bumped up, eating up some of the previously “free” portion of the block. The newly-reserved memory is then usable for the reference type instance that has just been allocated. That is extremely cheap; just a single pointer move, plus zeroing out the newly reserved memory if necessary.