Related posts
Tomasz Tunguz
x.com
The better the harness, the better the business. Berkeley published a study this week showing harnesses, the systems that control AI agents, set the price of an answer. The right harness cuts the cost of the same result by 71% without a loss of accuracy.
berkeley harnesses cuts
The the words it uses that this site has seen
least often elsewhere. Posts are matched on those words alone —
nothing here is a summary of this one.
Sources
All Korrents Recommends Blog x.com Mastodon Newsletter Bluesky Changed their mind Site
Top people
Sebastian Raschka
Bill Gates
Brad DeLong
Armin Ronacher
Brian Potter
Oliver W. Kim
Tomasz Tunguz
Gergely Orosz
Brad Feld
Sanjay Bakshi
François Chollet
Addy Osmani
Showing Profile →
Show everything
Hiding Show them again
Show them again
Further back ↓
Hiding Show them again
19 September
Economics professor at the University of California, Berkeley and a former Deputy Assistant Secretary of the US Treasury. Author of Slouching Towards Utopia and writer of the Grasping Reality newsletter.
Their words
The Reagan Revolution promised faster growth in exchange for lower taxes on the rich. The rich won; the growth never showed up.
Show the whole quote
braddelong.substack.com
18 September
Creator of Flask and Jinja. Writes about software at lucumr.pocoo.org.
Their words
most harnesses need some pruning on compaction anyways for cost reasons which Jev might help with
Show the whole quote
@mitsuhiko on X x.com
Creator of Flask and Jinja. Writes about software at lucumr.pocoo.org.
Their words
I would not dismiss the idea of using Jev for compaction at all. First of all because most harnesses need some pruning on compaction anyways for cost reasons which Jev might help with. I can see this being quite interesting.
Show the whole quote
x.com
17 September
Venture capitalist and founder of Theory Ventures, previously a partner at Redpoint. Writes a near-daily blog on startups, SaaS metrics, data and AI.
The Harness Margin Opportunity A UC Berkeley study finds the harness sets the price of an answer : GPT-5.6 Sol costs 71% less on Pi than on Claude Code, the same model returning the same result, & none of the 42 harness comparisons shows a statistica…
Anthropic OpenAI Related
16 September
Writes The Pragmatic Engineer, the software-engineering newsletter, and wrote The Software Engineer's Guidebook. Formerly an engineering manager at Uber, and at Microsoft/Skype and Skyscanner before that.
Their words
vibe coding is not dead: Non-eng teams inside of tech companies are just getting started with this approach, building on top of internal “harnesses for non eng teams” They are also learning limitations of “vibe coding” v quickly!
Show the whole quote
@GergelyOrosz on X x.com
15 September
AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).
Some food for thought when designing benchmarks... So, here's a little computer-use (visual) comparison between GPT-5.6 Astra and Qwen3.8 Max. The task here was to recreate the image in the center using the Paint UI. Super interesting how the two different LLMs+Harnesses approached this totally differently by default. I.e., Astra tried to approach this by drawing and layering geometric shapes. Qwen approached this pixel by pixel. (Of course, the pixel-by-pixel result looks closer to the original, it's essentially a low-res version of that by nature.) So, the Qwen-generated image would surely… LLMs OpenAI Related
5 September
Economics professor at the University of California, Berkeley and a former Deputy Assistant Secretary of the US Treasury. Author of Slouching Towards Utopia and writer of the Grasping Reality newsletter.
I asked a local instantiation of google/gemma4:31b-mlx to introduce itself to my new research assistants and point at its own workspace. What came back was a meditation on Plato, salamander brains, and why you should treat it like an eccentric roommate: **ThothBot Makes an Appearance in My RA-Group Telegram Chat: FRIDAY FOG & FLATLANDS** Living & working in Berkeley, California: An AI writes an introduction to my research assistants, & PanGram/SubStack grades it 100% not-slop; or, what... 1/ Google Related
3 September
Co-founder of Foundry Group and of Techstars. Has blogged at Feld Thoughts since 2004 and has written or co-written nine non-fiction books on startups and venture capital.
Chapter 24.1: The Third Row Eleven people know there are three founders, not two. Their lawyer cuts every promise. These are the people trying to save her. "She's twenty-nine. It was thirty-one degrees in Leadville last night." "Then find her." zeroknowledge.ink
startups Related
Indian value investor who taught Behavioural Finance and Business Valuation as an adjunct professor at MDI Gurgaon. He writes the Fundoo Professor blog about investing, moats and mental models.
Apple's market cap was about $351–360 billion when Steve Jobs died. On September 1, 2026, Tim Cook's last day as CEO, it was $4.75 trillion. When Ballmer handed over to Satya Nadella in Feb 2014, Microsoft market cap was about $315B. It’s $3.7 trillion now. Similar outcome with Sam Walton and Wal-Mart. It went the other way with GE and Jack Welch. So, it cuts both ways, and I think the deciding factor is whether the underlying business quality, moat, and capital-allocation discipline survive the transition intact. Quoting @menakadoshi The events at HDFC Bank confirm two of my worst suspicions. One, celebrity founders and CEOs often leave behind a leadership vacuum. As the adage goes, nothing grows under a banyan tree. India is witnessing that more frequently as part of the great handover from post-liberalization leaders to the n… Related
Creator of the Keras deep-learning library and the ARC-AGI benchmark.
Their words
so harness capabilities are increasingly shifting into the model itself.
Show the whole quote
@fchollet on X x.com
19 August
Engineering leader on Google Chrome; writes about web performance and, lately, about working with AI coding tools.
Their words
One of the things I found most exciting in the last couple of years was seeing as model qualities gotten better and and harnesses and tools have gotten better, how many people um that were directors or VPs or SVPs or any of these levels were actually rolling up their sleeves and trying things out.
Show the whole quote
youtube.com
18 August
Indian software engineer and co-founder of Skcript; builds web apps and writes at varunraj.in.
Long power cuts on every alternative days, What is @TANGEDCO_Offcl doing? That too after a recent power cable change for the entire street. TNEB scam is still hapening in @TVKPartyHQ government? government Related
12 August
Co-founder and CTO of Honeycomb; previously an infrastructure engineer at Parse, Facebook and Linden Lab, and co-author of Observability Engineering.
Their words
And it wasn't actually the models. It was the harnesses. It was all the tooling.
Show the whole quote
youtube.com
30 July
AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).
Consistent with what I found with Qwen3.6 a while back: Claude Code uses 2-3x as many tokens than (many) other harnesses at similar success rate. - Unoptimized? - Buggy? - Deliberate (coz that helps it in more challenging cases)? Need to find some time to investigate more... Quoting @composio We ran Kimi K3 through 3 agent harnesses (Claude Code, Hermes, Kimi Code) on 28 identical tasks. All 3 harnesses completed the tasks at similar success rates, but the interesting story is token efficiency: the same task cost up to 30x more tokens depending on the harness. 🧵🧵 Anthropic Related
18 July
Structural engineer and senior infrastructure fellow at the Institute for Progress; writes Construction Physics on why physical things cost what they cost.
Reading List 07/18/26 A setback for California Forever, a $500 million artillery shell plant not producing any shells, the difficulty of recycling EV batteries, manufacturing wire harnesses, and more.
Related
15 July
Investor and writer. Previously a partner at Andreessen Horowitz and a product leader at Twitter, Facebook, Snap and Microsoft. Writes essays and memos at sriramk.com.
14 July
Co-founder of Modem and a founding engineer at Sentry, where he went on to be VP of Engineering. Co-author of Third-party JavaScript, and co-host of the State of Agentic Coding podcast with Armin Ronacher.
🎙️ State of Agentic Coding #8 with guest Mario Zechner is out. In this episode: - Vibe-checking Fable & GLM 5.2 - @mitsuhiko.at teaches us RL - How harnesses smooth out model jank - Loops, surviving AI fomo, and more YT: www.youtube.com/watch?v=_lfp... Spotify: open.spotify.com
Related
12 July
Founder and CEO of Social Capital, a venture firm; co-host of the All-In podcast and an early Facebook executive. Writes an annual letter and a weekly newsletter on markets and technology.
27 June
AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).
21 June
Programmer; founded comma.ai and tinygrad, and was the first to unlock the iPhone.
6 June
AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).
15 May
Writes Very Serious, a newsletter on politics, the economy and culture, and hosts the podcast of the same name.
22 October 2025
Fellow at the Manhattan Institute writing on crime, drugs and policing; writes The Causal Fallacy.
10 October 2025
Writes on effective altruism, energy and the arithmetic behind arguments people repeat — including how much electricity a chatbot query actually uses.
30 August 2025
Structural engineer and senior infrastructure fellow at the Institute for Progress; writes Construction Physics on why physical things cost what they cost.
data centers America
Their words, before
Taking this into account, actual water consumed by data centers is around 66 million gallons per day.
Show the whole quote
Their words now
So to wrap up, I misread the Berkeley Report and significantly underestimated US data center water consumption.
Show the whole quote
26 July 2025
Co-founder of Microsoft and of the Gates Foundation. Writes Gates Notes on global health, energy and climate, and on whatever they have been reading.
@NickKristof's journalism inspired me to get involved in global health decades ago, and I hope his reporting on the devastating consequences of aid cuts inspires others to act today. Quoting @NickKristof One reason we tolerate the evisceration of USAID is skepticism about aid's effectiveness. Yet aid can and does save lives. As @BillGates notes here, child deaths from diarrhea have fallen 72% since 2000. Diarrhea still kills too many children, but we know to solve these problems, inexpensively. We… Related
11 July 2025
Co-founder of Microsoft and of the Gates Foundation. Writes Gates Notes on global health, energy and climate, and on whatever they have been reading.
The devastating effects of these cuts are entirely preventable—and it’s not too late to reverse them. Quoting @samstein An HIV doctor in Africa, whose work depends on USAID and PEPFAR, sends in a dispatch of the bleak situation now unfolding there Related
3 July 2025
Co-founder of Microsoft and of the Gates Foundation. Writes Gates Notes on global health, energy and climate, and on whatever they have been reading.
The facts are simple and devastating: Aid cuts have already cost lives, and the number of deaths will continue to rise. Here’s the evidence. Related
19 May 2025
Founder of what became Bloomberg New Energy Finance; writes and speaks on clean energy, hydrogen and the economics of the transition, and hosts the Cleaning Up podcast.
3 April 2025
Brazilian software engineer, creator and maintainer of GoReleaser; works on Copilot CLI at GitHub. Writes about tooling and his setup at carlosbecker.com.
22 March 2022
Engineer writing about the physical economy — oil and gas, manufacturing, batteries, drones and what it costs to build things.
Their words
Trucks, vans, and other high mileage vehicles promise to impact oil demand more than EVs.
Show the whole quote
austinvernon.site
26 July 2019
Development economist; writes Global Developments, on growth, history and the numbers behind them.
The RA’s Guide to Busia Spending a summer doing fieldwork in Busia, Kenya has become a rite of passage for Berkeley development RAs. When I was planning my trip, I wish there’d been a guide with the collected experience of former RAs all in on…
Related
7 February 2019
Development economist; writes Global Developments, on growth, history and the numbers behind them.
Welcome! Hello. I am a first-year PhD student in Economics at UC Berkeley, starting in 2018. I did my undergrad at Harvard College, graduating in 2016, and worked for two years at the New York Fed.
Related
9 March 2018
Associate Professor at MIT CSAIL, leading the Programming Systems Group; works on programming systems that manipulate uncertainty, from neural networks to unreliable hardware.
Their words
Neural network pruning techniques can reduce the parameter counts of trained networks by over 90%, decreasing storage requirements and improving computational performance of inference without compromising accuracy.
Show the whole quote
arxiv.org
Chief AI Scientist at Databricks, after its acquisition of MosaicML, where he was a founding team member; known for the Lottery Ticket Hypothesis from his MIT PhD (ICLR 2019 Best Paper).
Their words
Neural network pruning techniques can reduce the parameter counts of trained networks by over 90%, decreasing storage requirements and improving computational performance of inference without compromising accuracy.
Show the whole quote
arxiv.org
4 January 2017
Game developer and moral philosopher; made QWOP, GIRP and Getting Over It, and teaches at the NYU Game Center.
Some local-multiplayer deep cuts In developing the repertoire of games for Multibowl, I played about 1000 games made from 1975 through to about 2000. I found a few gems that are worth playing in their own right, and which I think aren’t all that well k…
Related
Nothing matches. Show everything
A korrent is a belief a person has stated in their own words: one
sentence stating the claim, backed by a quote and a source, kept at
korrents.com .
Under a name here, the quoted block is what they actually said.
The korrent beneath it is the claim those words support, in
korrents' wording — tap it to see the record, its source, and who
else holds it.
Nobody here wrote their own korrents. They are compiled from public
statements, and a person can change their mind, which is recorded too.
Got it
About the English under a post
Some people here publish in a language other than English. Where they
do, this site shows a machine translation beneath the post, in
this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the
serif above is exactly what the person published, and it is what to
quote them on. A translation can be wrong in ways that matter,
especially about tone.
Only the post's own words are translated. A quoted post, a linked
article and a belief on korrents.com
are left in their original language.
Got it