Julio Castro

In the relentless pursuit of excellence

Entrepreneur | Education | Tech | Finance | Marketing

[Heron Intelligence @Heronintel]

$IBM pre-announced a miss that cost the stock 25%, its worst day in 110 years. Management blamed timing: clients shifted capex to AI hardware and deals slipped. None of the four former operators we spoke with after the print accepted that whole.

A former executive who spent over a decade in IBM operations rejected the budget-reallocation narrative, arguing the same shift never hit other large integrators.

A senior technology executive with three decades inside called tens of deals slipping at once alarming and possibly multi-quarter. In Q1, a former VP in operations told us he would challenge the mainframe-flywheel assumption and questioned whether Z growth could hold three more quarters. One quarter later, Z fell 42%. "This never came up when this was impacting, let's say, the Accentures of the world or the Cognizants of the world, and IBM was winning in that market. Suddenly, in July 2026, you are hearing about the client budget reallocation, which is not affecting other players, but it's just affecting IBM."

The sources who engaged expected the slipped deals to return renegotiated, not at held prices. A former director in commercial operations who worked on ELA deal mechanics described clients using the slip to renegotiate scope and pricing, knowing IBM needs the close. A senior technology executive expected bundled products or service credits with the same result: diluted margin. "I have seen customers using, let's say, this slip to definitely renegotiate scope and pricing, especially in large deals. Because clearly, they know that IBM has the financial strength to support and negotiate it down or a discounted price."

Three of four described the Red Hat, HashiCorp, and Confluent engine as real, with IBM strongest where regulation keeps workloads private. The contested question is whether it compounds. Sources pointed to flat consulting as evidence the gen AI book of business is not converting, the same read our Q1 calls surfaced when a former VP said AI revenue was unreliable and consulting should rise or fall with software.

The observable forward: whether remaining deals close early in Q3 or stretch late, and whether the terms preserve the margin the FCF guide assumes.

Source views are those of former employees interviewed by Heron. Not investment advice.

[Julio Castro @jccastro94]

@ThrDataSnob @jeremyphoward As a fountain pen enthusiast, $3 pens sounds pretty good to me 🤣 Fountain pens probably wouldnt work in space but I would for sure take one up there either way.

[Julio Castro @jccastro94]

@mrdbourke Cant find the article. This is a bit more recent: https://t.co/q7ql8Va8tl
And it seems to be possible, but it will depend on your mobo and cpu as well.

[Julio Castro @jccastro94]

On consumer hardware? I think it depends on the motherboard mostly. And usually, if you are using 2 gpus, the lanes are split and less efficient so you cant use them to their fullest potential.

I’ll try to find the article on it, but last I saw (when the 4090 came out) having to 4090s together on the best consumer motherboard would give you the exact same 24gb of vram cause the first lane would only gove you 16 and the second 8.

[Anand Sanwal @asanwal]

Throwback Wednesday

Who remembers the daily deals space?

@Groupon anyone?

Today, one of the other early entrants to that space shut down

Just 7 months after being acquired by PE firm Regent, a firm that specializes in complex corporate divestitures, @zulily called it quits

The company had a ~15 year run (ultimately expanding beyond its daily deal roots)

* It first raised in 2009
* Then went public 2013
* Then was acquired by Qurate (parent co of QVC)
* And then acquired by PE firm Regent earlier this year
* And now dead

The last man standing in the daily deal space is now Groupon (although it is a shell of its former self)

Once valued at $12.7B and regarded as the fastest growing tech company of all time, it is now worth < $400 million

It's main competitor LivingSocial was valued at $6B before Groupon ultimately acquired it for nothing

Daily deals was a space much like Amazon aggregators more recently.

It got hot, had lots of capital flow in, had little to no real barriers to entry and so the result ultimately was no big sustainable businesses were built

Source - @CBinsights

Zulily company profile
https://t.co/znCobTjjv6

Groupon company profile https://t.co/Bh0GeicON1

LivingSocial company profile
https://t.co/YVdVCzktEA

[Julio Castro @jccastro94]

I'm going through your code and got a few questions - I know it's a demo but just want to get an idea of your reasoning.

1. Have you tried training with the full dataset? Other sets?
2. When prepping the data, shouldn't you delete all \n before creating the train.txt loop? Since there are linebreaks in the data, they would create new lines in the train.txt file and mess it up.
3. Would it be useful to limit the context? At what point does the preceding words become irrelevant to the newest word?
4. is __label__ necessary to label? and is the double space enough to distinguish input/output? (I've never used FastText so no idea)
5. Would love to hear you expand on your tweet "from serving as a tokenizer of a larger model, as an auxiliary model predicting possibilities to reduce the search at the time of inference, as a map for diffusion models prediction, or as a quick model"

[AK @_akhaliq]

Anydoor demo is out

Zero-shot Object-level Image Customization

demo: https://t.co/wVo8MpwVN8

a diffusion-based image generator with the power to teleport target objects to new scenes at user-specified locations in a harmonious way

[Mayo Oshin @mayowaoshin]

I was recently invited as a generative AI Keynote speaker at the Private Markets European Summit.

My presentation was focused on answering one key question: How can financial institutions integrate generative AI to gain a competitive edge whilst minimising risk.

Here are some key takeaways:

- AI can accurately analyse financial reports that contain tables and images using multimodal capabilities, and also provide references to relevant sections of the document.
- Deal sourcing and due diligence can be conducted utilising "agents" that search the web for data, integrate with proprietary documents, APIs and databases in parallel.
- Sentiment analysis can be used to analyse product reviews, comments, and brand strength of portfolio companies.
- Development tools like @llama_index and @langchain can make it easier to build a proof of concept AI app for stakeholder buy-in.
- Key generative AI risks include hallucination, IP infringement, data theft + poisoning, bias, and regulatory compliance.
- Risk mitigation strategies include "human in the loop", threat modelling, tool use + reranking, open source models, non-proprietary use cases i.e. customer support.

Finally, a key ingredient for integrating AI in an organisation is to hire AI expertise to design an effective strategy and technical architecture that guides your team on the "what" and "how", in order to successfully build a transformative AI solution.

If you're interested in working with me as this "key ingredient" to help integrate AI and transform your business, fill out this brief form: https://t.co/sL0IZQ2iAl

[shyamal @shyamalanadkat]

shipping AI without calling it AI is underappreciated. it brings back focus on users and value; no hype, just utility that users gradually rely on. often the best solutions hide complexity behind simple interfaces.

[Auren Hoffman @auren]

@DavidSacks Twitter has changed the core product a lot since Elon took over. The product iteration cycles are great

But the ad product is still pretty lousy and has not changed much in the last year. They need to think of the SMB advertiser as a customer and give them better tools.

[Julio Castro @jccastro94]

I’ve researched, dove in, and even volunteered at charities I donate to - and to a degree I agree that most charities are not fully transparent. But the issue with EA is that it’s become a “trend” and many are blindly following some influencers “research” instead.

EA as a concept is something I agree with, but it feels like in practice it’s not possible. In the end, you just gotta donate where you believe in the cause and then hope for the best.

[Julio Castro @jccastro94]

“There is no such thing as failure, only feedback”

[Sam Altman @sama]

Move faster. Slowness anywhere justifies slowness everywhere.

2021 instead of 2022. This week instead of next week. Today instead of tomorrow.

Moving fast compounds so much more than people realize.

[Sam Altman @sama]

Move faster. Slowness anywhere justifies slowness everywhere.

2021 instead of 2022. This week instead of next week. Today instead of tomorrow.

Moving fast compounds so much more than people realize.

[Lex Fridman @lexfridman]

I'm sitting here, experiencing my own little ups &amp; downs of life, and suddenly remember there's 8 billion other humans like me, now also living through their ups &amp; downs: from horrible lows to beautiful highs. What a wild ride we're all on... And we're on it together. Hang on ❤

[Auren Hoffman @auren]

One of the most repeated pieces of advice on World of DaaS: startups are so much harder than you think

it's one thing to know that, another to really understand it

[Peter Yang @petergyang]

It's only the incumbents who have proprietary data who are safe from OpenAI's rapid product velocity.

[Julio Castro @jccastro94]

@musdotdigital @OpenAI Hey, just found this randomly - quick Q, is it for apple podcasts? There is no validation and comments in the https://t.co/9Mft6Btquc mentions apple. Just wanna know what type of podcasts it can download, thnx

[Marc Andreessen 🇺🇸 @pmarca]

THE TECHNO-OPTIMIST MANIFESTO part 1

“You live in a deranged age — more deranged than usual, because despite great scientific and technological advances, man has not the faintest idea of who he is or what he is doing.”
— Walker Percy

“Our species is 300,000 years old. For the first 290,000 years, we were foragers, subsisting in a way that’s still observable among the Bushmen of the Kalahari and the Sentinelese of the Andaman Islands. Even after Homo Sapiens embraced agriculture, progress was painfully slow. A person born in Sumer in 4,000BC would find the resources, work, and technology available in England at the time of the Norman Conquest or in the Aztec Empire at the time of Columbus quite familiar. Then, beginning in the 18th Century, many people’s standard of living skyrocketed. What brought about this dramatic improvement, and why?”
— Marian Tupy

“There’s a way to do it better. Find it.”
— Thomas Edison

Lies

We are being lied to.

We are told that technology takes our jobs, reduces our wages, increases inequality, threatens our health, ruins the environment, degrades our society, corrupts our children, impairs our humanity, threatens our future, and is ever on the verge of ruining everything.

We are told to be angry, bitter, and resentful about technology.

We are told to be pessimistic.

The myth of Prometheus – in various updated forms like Frankenstein, Oppenheimer, and Terminator – haunts our nightmares.

We are told to denounce our birthright – our intelligence, our control over nature, our ability to build a better world.

We are told to be miserable about the future.

[Julio Castro @jccastro94]

Sometimes, you just need some time to reflect on how amazing the world we live in is. Take some time to explore nature, and I promise you'll come out better than you went in.

#CostaRica #PuntaUvita #MarinoBallena #Nature #Ocean