David Olinsky (@dolinsky) profile picture

Must build fun stuff. Director of Enterprise Architecture @sapp_digital. Pet lover. Python/Data geek. Deadpool @decantr.

I miss Hoboken, NJ
Joined April 2008
Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)Photo by David Olinsky (@dolinsky)
Last quarter I rolled out Microsoft Copilot to 4,000 employees. $30 per seat per month. $1.4 million annually. I called it "digital transformation." The board loved that phrase. They approved it in eleven minutes. No one asked what it would actually do. Including me. I told everyone it would "10x productivity." That's not a real number. But it sounds like one. HR asked how we'd measure the 10x. I said we'd "leverage analytics dashboards." They stopped asking. Three months later I checked the usage reports. 47 people had opened it. 12 had used it more than once. One of them was me. I used it to summarize an email I could have read in 30 seconds. It took 45 seconds. Plus the time it took to fix the hallucinations. But I called it a "pilot success." Success means the pilot didn't visibly fail. The CFO asked about ROI. I showed him a graph. The graph went up and to the right. It measured "AI enablement." I made that metric up. He nodded approvingly. We're "AI-enabled" now. I don't know what that means. But it's in our investor deck. A senior developer asked why we didn't use Claude or ChatGPT. I said we needed "enterprise-grade security." He asked what that meant. I said "compliance." He asked which compliance. I said "all of them." He looked skeptical. I scheduled him for a "career development conversation." He stopped asking questions. Microsoft sent a case study team. They wanted to feature us as a success story. I told them we "saved 40,000 hours." I calculated that number by multiplying employees by a number I made up. They didn't verify it. They never do. Now we're on Microsoft's website. "Global enterprise achieves 40,000 hours of productivity gains with Copilot." The CEO shared it on LinkedIn. He got 3,000 likes. He's never used Copilot. None of the executives have. We have an exemption. "Strategic focus requires minimal digital distraction." I wrote that policy. The licenses renew next month. I'm requesting an expansion. 5,000 more seats. We haven't used the first 4,000. But this time we'll "drive adoption." Adoption means mandatory training. Training means a 45-minute webinar no one watches. But completion will be tracked. Completion is a metric. Metrics go in dashboards. Dashboards go in board presentations. Board presentations get me promoted. I'll be SVP by Q3. I still don't know what Copilot does. But I know what it's for. It's for showing we're "investing in AI." Investment means spending. Spending means commitment. Commitment means we're serious about the future. The future is whatever I say it is. As long as the graph goes up and to the right.
5,082
25,164
5,922
169,564
26,309,251
Today we are making the @orioledb Patent freely available to the Postgres community Oriole is a storage extension for Postgres which uses Postgres' pluggable storage system. It's a drop-in replacement for Postgres' existing storage engine (Heap). Our north star is to upstream Oriole to be part of the Postgres source tree, developed and maintained in the open alongside the rest of Postgres (if the community want it!) We announced the acquisition of Oriole over a year ago. Since then, we have been working on cleaning up the legal structure and finalizing the asset transfers. We have wrapped up all legal activities, and we fully own US Patent 10,325,030 (“Durable multiversion B+-tree”). @supabase is making this Patent explicitly available to all Oriole users (including proprietary forks) in accordance with the open source license. Oriole benchmarks show that it is around 5.5x faster than the default storage engine (Heap):
15
78
7
576
186,070
David Olinsky retweeted
This is the craziest AI news this month An open source generative physics engine able to generate 4D worlds
224
805
234
4,992
609,932
Look at this graph and explain to me why you think I should read this as a measure of consumer confidence rather than partisanship.
297
753
72
6,356
417,652
Hey Trump voters. Hate to break it to you - but here’s what actually happened on Tuesday. Watch. Bernie explained it 20 years ago.
2,015
18,912
2,091
84,430
6,298,633
David Olinsky retweeted
Keep your app tag lines simple. Big words like "Democracy is at stake" means nothing to people. Instead try "Eggs will cost less"
78
190
28
5,066
269,467
The man who posted the fake Russian video claiming Haitian illegals are voting in Georgia, AlphaFox78, was paid by the Russians. Look forward to those that shared the disinformation apologizing for helping our enemies.
205
2,310
72
6,022
588,202
David Olinsky retweeted
🧵Next week, you will see people using something called Benford's Law to try to prove election fraud. ⛔️These people are wrong⛔️ I am a scientist who has published on Benford's Law. Let me tell you what it is and why what they are doing is mathematically incorrect. 1/
97
1,584
200
6,503
736,987
Today I released my plan to help build a brighter future for Puerto Rico and the Puerto Rican people as president. Meanwhile, at Donald Trump’s rally, they’re calling Puerto Rico “a floating island of garbage.”
7,806
8,996
622
49,008
4,310,565
IDF just went It's a Mad, Mad, Mad, Mad World on Lebanon and told everyone where Hezbollah's millions of dollars in cash and gold are hidden. 🤣
1,279
7,988
936
48,873
4,362,132
11/11 So Trump and his team *in their own words*: -Knew they lost -Knew it was not legal for Pence to over turn -Knew the states did not have fraud issues -Researched "alternative electors" a month in advanced of those claims -Knew the documents were unofficial -Took part in the transfer process -And attempted to pass off forged documents directly to the VP -Because of a pre-meditated plan to steal the election. The first volumes represent 1889 pages of evidence, 70% still under seal. This is just from the 30% unsealed, and there are more volumes to go. The case could not be more clear. And as a reminder, Trump's team refused to dispute the authenticity of any of this evidence, they only ask for it to be delayed as it would "hurt their client" during the election.
244
3,854
403
12,014
417,040
An Elon Musk-funded PAC is microtargeting Muslim people in Michigan and Jewish people in Pennsylvania with opposite messaging about Kamala Harris using ZIP code targeting, according to data I pulled from Snapchat. Few things from this investigation: 404media.co/this-is-exactly-…
270
3,976
718
9,966
2,010,771
“We have been Republicans all our lives. We both voted for Donald Trump… Never in a million years did either of us think that we'd be standing here supporting a Democrat. But we've had enough. We've had enough.” #RepublicansForHarris
2,088
11,640
561
48,804
1,634,544
David Olinsky retweeted
"I'll start tomorrow" The biggest lie you tell yourself:
32
708
15
4,047
626,993
David Olinsky retweeted
The <100B Model Red Wedding I do not think that people who criticize OpenAI have sufficiently absorbed the magnitude of disruption that has just happened because of 4o mini Llama 3 70b: 82 MMLU, $0.90/mtok gpt 4o mini: 82 MMLU, $0.15/mtok every model on the RHS side of this chart is now strictly dominated by their LHS counterparts some of these models were SOTA 3 months ago. what is the depreciation rate on the FLOPs it took to train them? gpt4 took $500m to train and it lasted ~a year. intelligence too cheap to meter, but also too ephemeral to support >5 players doing R&D? is there an angle here i'm missing? the other angle i have been thinking a lot about is the separation of reasoning from knowledge. RAG/memory plugs knowledge easily but not reasoning. 82 MMLU is plenty. you can get it up to 90, but its not going to be appreciably smarter in normal use without advancing other metrics. so in 2025 we're likely to evolve towards 0) context utilization (RULER) 1) instruction following (IFEval) 2) function calling (Gorilla) 3) multistep reasoning (MUSR), 4) coding ability (SciCode), 5) vision understanding (VibeEval?) for all the stuff that RAG can't do.
[17 July 2024] Mini, Nemo, Turbo, Lite - Smol models go brrr! Vibe check of 4o vs mini for instruction following vs summarization: - we find some examples where mini is worse, some examples where it is better. Mini is subjectively: - unevenly worse at formatting instruction following - comparable at ranking/topic selection/grouping - comparable at subtitling - comparable at structured output - full manual eval with scores pending... - but for 3.5% of the price of 4o, the fact that we have to spend time deciding is shocking!!!
32
62
15
641
229,223
David Olinsky retweeted
It's like she's debating MAGA Twitter come to life. Victor Orban, dead pets, Ashli Babbitt, "J6." She's debating Catturd.
432
3,734
217
33,296
1,239,530
David Olinsky retweeted
The world is not “going to shit.” We live in the best time in human history by most measures.
Can anyone pinpoint the exact moment when the world started to go to shit?
440
1,022
192
5,876
733,535
I didn't have service during today's event here in Asheboro, NC, but former Pres. Trump's comments below on the significant downward jobs revisions released by BLS today were almost entirely - and intentionally - untrue. Egregiously so. I'll take them one by one.
196
1,678
175
7,492
1,639,410
David Olinsky retweeted
Transformer Explainer Really cool interactive tool to learn about the inner workings of a Transformer model. Apparently, it runs a GPT-2 instance locally in the user's browser and allows you to experiment with your own inputs. This is a nice tool to learn more about the different components inside the Transformer and the transformations that occur. Tool: poloclub.github.io/transform…
15
454
17
1,905
122,042
David Olinsky retweeted
In 2016, researchers at the University of Adelaide tested Kurt Vonnegut's theory that, "There’s no reason why the simple shapes of stories can’t be fed into computers." They took the emotional arcs of 1300+ novels from Project Gutenberg, turned that into data, used modern tech to analyze the emotional arcs, and then identified 6 patterns seen over and over again in western storytelling. Here they are: 1. Rags to Riches (rise) Your classic underdog tale. A humble, hardworking peasant climbs the mountain to pull the sword from the stone. • Rocky • King Arthur • The Pursuit of Happiness 2. Riches to Rags (fall) Maybe the saddest story of them all. A journey from the highest of highs to the lowest of lows. • King Lear • Citizen Kane • Scarlet Letter 3. Man in a Hole (fall then rise) A character’s doing fine, gets herself into a huge problem, but figures out how to overcome it. They often end up better than they started. “You see this story again and again,” Vonnegut says. “People love it, and it is not copyrighted.” • The Martian • The Hunger Games • Shawshank Redemption 4. Icarus (rise then fall) The hero goes on a meteoric rise up New York (or some other) society, calls everyone “old sport,” and throws the wildest parties in town. Then reality sets in, and he realizes he’s too close to the sun. • Macbeth • Great Gatsby • Death of a Salesman 5. Cinderella (rise then fall then rise) I’ll leave this description to Vonnegut: “We’re gonna start way down here. Worse than that, who is so low? It’s a little girl… the shoe fits, and she achieves off-scale happiness.” • Red Rising • Slumdog Millionaire • The Count of Monte Cristo This is my personal favorite. 6. Oedipus (fall then rise then fall) Up until the ~70% mark of the story it looks like things are sunshine and rainbows. Walter White goes from high school teacher to king of the drug lords, if you will. Then all goes wrong. The original fall is often not their doing while the final fall is. • Hamlet • Gone Girl • Breaking Bad My 3 takeaways: 1. Rags to Riches, Oedipus, and Cinderella rank as the three most popular with consumers. AKA, those books sold the most copies. 2. When you think through a story, give it an emotional shape. Literally draw it. X axis: Time Y axis: Ill fortune to good fortune You might be surprised how much it helps you craft your plot (I was shocked). 3. Vonnegut was a damn genius.
87
857
116
5,316
738,065