semessier

well it also influenced judgement:

the freshly Lou Gerstner Jr. who later danced the elephant into muddy waters appointed "Internet Division" head below

https://blog.irvingwb.com/blog/2005/11/the_ibm_interne.html

declared in meaning that Notes is for Enterprises and the web browsers for consumers/hobby. Now to settle this once and for all the IBM web browser was canceled.

I would parse it as event sourcing at its finest + cqrs

not sure about the use of exponential and efficiency frontier here, these have formal sides to them but seem to be used rather inflationary and colloquially.

cudos

Apple should be required to make RAM including notebooks customer-upgradeable/swappable including LPCAMM2.

for all the people posting founding in Ireland or Estonia (or England!!) etc. The local bureaucracies will spend infinite resources on nailing anyone down on the place of business being != to where the registration is. That is coercing one way or another to incorporate with the full overhead locally after all and most importantly that taxes are not happening in Ireland but to feed the local administration.

the real problems with administration dome starts after founding in many but not all European countries.

There is light at the end of the tunnel as an EU Inc. is proposed.

However the bureaucrats in probable but all countries will try to water it down to pointless to keep or extend their responsibility territory and duties.

this is an all-stock transaction. No cash spent.

and where is the Transformer library ;)

this is an empirical engineering paper with theoretical dressing, it would not need to be a theorem paper of course.

the GPU builds are very high stakes games of depreciation: if the mission life is e.g. 4 years you win, if a disrupting ASIC for the transformer comes in you lose.

As of today the gamblers seem to win, demand even for A100s, H100s is high prices are even rising.

this lacks the math for any bold claims

it would be really interesting when that moment was at probably OpenAI when they realized that this was doing more than next word prediction but signs of <you name it>

what could go wrong in the recursive loops running today 24/7 probably. Attended/unattended almost makes no difference any more, no human can grasp probably numerous changes per iteration. This is outright dangerous.

V being collinear is obvious, the question is/was also which additional orthogonal projections such as camera position for vision would improve the transformer.

the real interesting question would be if it then does its language-based reasoning also in short form and if so if quality is impacted.

it's not a new question if the as-is programming languages are optimal for LLMs: a language for LLM use would have to strongly typed. But that's about it for obvious requirements.

SIR_TONY_HOARE = μX • (think → create → give → X)

-- process ran from 1934 to 2026 -- terminated with SKIP -- no deadlock detected -- all assertions satisfied -- trace: ⟨ quicksort, hoare_logic, csp, monitors, -- dining_philosophers, knighthood, turing_award, -- billion_dollar_apology, structured_programming, -- unifying_theories, ... ⟩ -- trace length: ∞ The channel is closed. The process has terminated. The algebra endures.

unless its greatly exagerated - he was quite mind sharp in his 80s

SIR_TONY_HOARE = μX • (think → create → give → X)

-- process ran from 1934 to 2026 -- terminated with SKIP -- no deadlock detected -- all assertions satisfied -- trace: ⟨ quicksort, hoare_logic, csp, monitors, -- dining_philosophers, knighthood, turing_award, -- billion_dollar_apology, structured_programming, -- unifying_theories, ... ⟩ -- trace length: ∞ The channel is closed. The process has terminated. The algebra endures.

looks like he is trying to make a point that the actual (formal) proof for 2Z + 1 (odd numbers) is still human - by himself that is. Not sure who came up with the core modular arithmetic idea of with s = 0 k increasing by 2 mod m.

that's pretty shallow for the front page. What would be interesting in this context are things such MXFP4 quantization etc. not commonplaces.

that's not a proof

a parameterization would be helpful for machine designers probably

He was Right on outsourcing, wrong on everything else.

Gerstner joined IBM in 1993 with the company in a very bad situation and immediately killed the "Baby Blues" breakup plan - no more independent product units for PCs, various servers, storage, semiconductors. Instead: one monolith optimized for selling professional services to enterprises. Outsourcing became the flagship.

This saved IBM then. It also set a ceiling on what IBM as a technology company could ever be again. The services pivot worked commercially. Outsourcing grew into a maybe 20 billion/year business, ironically eventually spun off as Kyndryl in 2021. But the cultural shift - from building systems/cutting edge-tech to managing other people's systems - hollowed out IBM's product DNA.

Consider that IBM's last major hardware win was the RS/6000, circa 1990. That's 35 years without a new product category victory. Meanwhile, IBM had countless pieces to own which is cumbersome -- and sad -- to list, and the list would be long. VMs invented in the 1970, running on every mainframe as an example: VMware ate the corporate market while the cloud hyperscalers built actual technology services based on it. You could probably get business consulting on it from IBM in the same time.

IBM Research remained well-funded throughout - still producing papers, still winning awards. But research was not focused to support a technology business any more, it even got pulled into service client engagements (!), subordinated to billable hours rather than productization. The research labs that gave us the bar code, DRAM, RISC, relational databases, and the scanning tunneling microscope became a presales asset.

The counterexample of large technology company in later trouble is Siemens: a conglomerate that actively manages distinct businesses as distinct businesses. Healthineers spun out and thrived. Energy spun out when it needed different go-to-market models. Industrial automation runs independently. No insistence on "one face to the enterprise customer", yet they can probably provided it if needed (don't know). They did their portfolio reshaping without strategic rigidity and the Gerstner betting on a single non-really-tech managed IT services anchored in shallowish business more than anything else horse and not the technology herd of horses.

Gerstner's legacy is complicated. He unquestionably prevented IBM's collapse financially then. But the rescue vehicle became a prison. A company with IBM's research depth, enterprise relationships, and installed base should have built the cloud, should have owned enterprise AI, should have been where AWS, VMware, TSMC even, and more low-hanging Salesforce are now.

Gerstner himself probably wouldn't be recognized as an IBMer by most who worked there before him. Brought in by a probably desperate board to save a company he then remade in a different image entirely. He saved IBM by making it something else he knew to be valuable from his prior work as a business leader. Whether that something else was the best available option, or merely an option he was familiar with: purchasing abstract solutions from a single source, is the open question his tenure leaves behind. This post argues it was the wrong one, dead wrong that is. In hindsight it was just good for the business strategic outsourcing (which is managed IT for large enterprises) side of things, wrong for much else.

Siemens as a company with similar issues of a collapsing core business ca. 2000, even much more so than the mainframe ever did, is the anti-thesis in terms of approach taken. It actively manage distinct businesses as distinct businesses, reshaping portfolios without insisting on a single face or a single model; an approach that turns out to work better looking at it now in comparison than trying to play seller of a single face to enterprise customer as a monolith. Most of the business are both technologically strong and financially in a good shape.

IBM under Gerstner set course on being a monolith optimized for selling and delivering professional services, stripped of the technology capabilities the company was built on. IBM evidently was saved then. Was it the only/best route to take? Probably not, as the low ceiling on what IBM as a technology company could be afterward turned out to be too low to stand or produce not a single major hardware win thereafter for over 35 years.

More bluntly: he transformed a legend into a fallen angel. IBM today is a shadow of what it was, what it stood for, and what it could have been. Gerstner himself fits the pattern of a business turnaround guy - an outside business consultant who saved IBM by remaking it in his own image. He never became an IBMer. But he did make IBM something else.

triggering the cold signal artificially might be an approach

> No standard has ever been developed using money obtained by selling copies of the standard.

unfortunately there are examples in the Telecom world

11 billion invested in multiple hardware developments including moonshots would have been more sensible that for Kafka operator (remember MQSeries/MQ btw)

the price sounds a little bit high from a technical perspective

without having implemented inference, just by looking at it from a math perspective this is base linear algebra/BLAS. I am very much wondering what a lean inference optimized API with covering 80% of all use cases across dtypes and sparsity would look like. Probably a far cry from what's in CUDA and probably all that's needed for practical inference.

Hence lossless does not seem plausible