Adam Fish

Seeing a lot of these tools which excites me. Local first via edge AI is starting to take root. Outside an algorithmic LLM breakthrough the models are asymptotic meanwhile hardware continues to improve. This shifts power back to end users and edge devices!

@WillManidis Agree been wanting someone to just test this concept of the injection act itself is the medicine

that taking supplements wasn’t close enough to what our brains perceive as the “practice of medicine” and therefore were mostly ineffective for things like back pain that were “mostly in your head” in a Sarno frame. but the needle is very clearly a clinical act

i have a very woo-woo view that a large percent of the population is suffering from sarno-described pain and that peptides are serving as a totemic fix to this because the most powerful placebo is one that requires physically injecting yourself and related power structure

Started writing @dittolive in my apartment ~8 years ago. This week I danced down Canal Street with a brass band and 120 teammates. We make operations unstoppable. Building the company behind that mission is mostly hard. Weeks like this remind you why you do it.

HoloStore is about 3 weeks of Codex. Static membership. Static partitioning. Static replication. Now that I have 9 diff linearizability & correctness tests I'm pushing it harder Tonight I told Codex to tackle CockroachDB inspired cluster membership & range rebalancing 5h later

Ditto is already playing a role in driving edge AI. Excited as this grows!

@lizvstein @AndyLapsa @clattner_llvm @marcberte_ceo @SRFtweets Congrats!! 🎉

Such a fun thread and hope it's a trend that continues. (Just had a baby myself and adore the generous portfolio company swag collection. Thanks to @AndyLapsa @clattner_llvm @Adam_Fish @marcberte_ceo @SRFtweets for all the love!)

We’re entering the Data Disintermediation Era. The most valuable startups are not replacing systems of record. They are intercepting data before it ever reaches them and owning the workflows that follow. The last two years were defined by AI voice and scribe companies. Not because transcription was novel, but because they captured high-value data upstream and used it to trigger billing, customer acquisition, and other revenue-critical workflows. The next phase will be even more powerful. The next two years will be defined by physical-world data disintermediation. Fixed or mobile/portable hardware will capture proprietary data at the source, where switching costs are real and ROI is directly measured against labor, error rates, revenue boosting activities, and compliance risk. Software can be replaced by the core SoR rolling out competitive products, but Hardware-anchored data pipelines will be challenging to reach to for incumbents.

Get the tech spec, and read all the lessons we learned along the way on our blog: builders.ramp.com/post/why-we-bu…

1/ Context graphs don’t really exist out in the wild today because they require joins across coordinate systems that don't share keys. Traditional databases solved joins decades ago. You have a customer_id, an order_id, a foreign key relationship. The join is discrete, the keys are stable, the operation is well-defined. Organizational reasoning requires a different kind of join. You need to connect: what happened (events) to when it happened (timeline) to what it means (semantics) to who owned it (attribution) to what it caused (outcome). These are five different coordinate systems. None of them share a primary key. And the keys themselves are fluid. "Jaya Gupta" in an email, "J. Gupta" in a contract, "@JayaGup10" in Slack. Same entity, no shared identifier. The join condition isn't equality. It's probabilistic resolution across representations in latent space Every existing data system optimizes for joins within a single coordinate space. Context graphs require joins across all of them simultaneously.

he is back github.com/bellard/mquick…

The new 1 Trillion parameter Kimi K2 Thinking model runs well on 2 M3 Ultras in its native format - no loss in quality! The model was quantization aware trained (qat) at int4. Here it generated ~3500 tokens at 15 toks/sec using pipeline-parallelism in mlx-lm:

@markgurman As a SME helping the EU, Apple won’t and can’t do anything: ditto.com/blog/cross-pla…

Apple is backed into a corner here. On one hand, they want to kill this ASAP. On the other, they don’t want another RCS-level PR and regulator war on their hands. Extremely curious what happens next. I’d lean toward Apple just blocking it and not caring.

Very cool demo! They have an incredible new product leader with @okdistribute

Brilliant technical deep dive from @Adam_Fish of @DittoLive into what comes after conflict-free replication, how CRDTs solve conflicts but not query consistency at scale, and a shift to systems that simply converge.

It’s true. The original reason for ending prices in .99 was not psychological, it was to make sure employees weren't stealing... and it starts with the cash register itself: In the 1870s, a saloon owner in Dayton was fed up with his bartenders pocketing cash and was desperate for a solution. While traveling by steamboat, he noted the mechanical counter that tracked how many times the ship’s propeller turned. He realized the same principle could track cash transactions... every sale increments a count, just like every propeller revolution. That insight led to his invention: "Ritty’s Incorruptible Cashier." Over time, they added a drawer and a bell and it became known as a Cash Register... and the name of the company became "National Cash Register" which you probably know as NCR, the company that still exists today (recently split into 2, worth a combined $4.5B) Every time a clerk rang up a sale, the drawer opened and the bell chimed. That sound told the owner money was (or wasn’t) going into the till. Merchants discovered that if they priced items at 49¢ or 99¢, the clerk had to open the drawer to make change, forcing the bell to ring and the sale to be recorded. Around this time in the late 1800s, prices began ending this way. The notion that $.99 feels cheaper than $1.00 didn’t come until much later. Marketing psychology picked up on it 50 years later, when professors started using terms like “left-digit anchoring” and consumer behavior reframed .99 pricing as a sales tactic rather than just an anti-theft device. NCR gave us more than cash registers... the company built modern sales culture and was a huge talent magnet, like the Salesforce and Ramp of its day. They created a formal sales school in 1887 to sell cash registers, complete with scripts, objection manuals, territories, quotas, and pep talks. They called it the “West Point of American business,” because it was known to be a talent magnet, producing executives who spread these practices across corporate America. In 1912, NCR was indicted for antitrust violations. Their star salesman was Thomas J. Watson, who was convicted and then fired from NCR. He took NCR’s sales playbook to a small company called the "Computing-Tabulating-Recording Company" and became its CEO. He took the NCR sales model into beast mode - expanding internationally and changing the name of the company to "International Business Machines" (IBM) to reflect that. A bunch of NCR disciples ended up at other prominent places, like Kettering at GM. Kind of wild that after 155 years they are still selling cash registers. anyway long way of saying to the haters in the comments that this poster is correct.

Creepy, gotta implement local first!

Mark - “the irony is that the Wi-Fi failed after years of making advanced technology”

@abstractalgo @localfirstconf We are actually moving over to QUIC in a future release vs. TCP today for performance and other benefits, so yes it will be!

@localfirstconf @Adam_Fish @Adam_Fish I just learned about multipath QUIC. is this compatible with Ditto? youtube.com/watch?v=XdSh1K…

"Why Local First matters to businesses" @Adam_Fish delivering a great talk on how local-first is transforming how businesses operate, improving revenue through higher uptime, streamlining operations by reducing dependencies.