what would a benchmark for communication clarity look like beyond short-horizon utterances like here?
I don’t think it’s impossible to bitter-lesson our way into a theory of mind (really, I don’t)
I don’t think it’s impossible to bitter-lesson our way into a theory of mind (really, I don’t)
Why is AI slop so frustrating to read?
AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.
Many of these slop patterns violate writing principles established in English writing books.
In our benchmark, AI writing far more often than human writing, with Claude Code replies by far the worst offenders.
We show examples of the most common slop patterns in this thread
0
0
0
103
There's so much potential for growth in inoculating yourself early from feeling schadenfreude towards others
0
0
2
139
SIMD Crawford 🟣 retweeted
2/ Shortest paper summary of the year: “Please don’t waste my time.”
The appendices, according to the review, were the authors’ original research proposal, still promising experiments that never ran.
The appendices, according to the review, were the authors’ original research proposal, still promising experiments that never ran.
1
3
97
8,266
RIP Richard Lloyd.
I spent many human-days trying to understand the lightness of his playing. Still one of my biggest motivations to keep playing.
music.youtube.com/watch?v=_TN3wB…
I spent many human-days trying to understand the lightness of his playing. Still one of my biggest motivations to keep playing.
music.youtube.com/watch?v=_TN3wB…
0
0
5
587
SIMD Crawford 🟣 retweeted
"god when is meta ever going to launch muse voice they announced it forever ago"
muse:
muse:
279
217
5,278
579,311
SIMD Crawford 🟣 retweeted
Come work with me at the Muse VR team. We’re hiring ML and RL people.
Superintelligence hits different when it’s on your face. 😎
DMs open
Superintelligence hits different when it’s on your face. 😎
DMs open
My team is hiring ML Engineers / Scientists.
Strong applicants have excelled in at least two disciplines in the past, low level and ML related, very good knowledge of current ML trends and classic ML/CV; and have 8+ years of experience (PhD may count for some).
DMs open.
0
3
49
3,119
SIMD Crawford 🟣 retweeted
Eisenkot getting ready to build a coalition
To all our friends in the United States and around the world:
The Israeli people have voted for change and to replace the current government.
Our democracy remains strong and vibrant. We are working to fulfill the will of the Israeli people by forming a new government as soon as possible - a national emergency government.
I intend to bring new leadership to Israel, leadership that serves all Israelis. Together with my partners, we will build a stronger Israel: A Jewish and democratic Israel. An Israel that is committed to liberal values, equality for all and is confident in its future.
The bond between Israel and the United States is more than an alliance between governments. It is a partnership between peoples, built over generations, sustained by shared democratic values, and strengthened by the conviction that our nations are more secure and more prosperous when we stand together.
Under my leadership, we will renew and deepen that partnership, strengthen the bipartisan foundations that have long sustained it, and work with our allies to advance a Middle East defined not simply by the absence of conflict, but by the presence of opportunity and cooperation.
Israel will remain one of America’s most reliable and capable strategic partners in the Middle East, a partnership that is mutually beneficial for both peoples, in every aspect.
I look forward to working closely and constructively with President Trump and his administration, whose friendship and commitment to Israel we deeply value. Together, we can advance our shared security interests, strengthen regional stability, and open new doors to prosperity for the peoples of the Middle East.
That is the Israel I believe in. And that is the future we intend to build.
לחברינו ושותפינו בארצות הברית וברחבי העולם:
אזרחי ישראל הצביעו בעד שינוי ובעד החלפתה של הממשלה הנוכחית.
הדמוקרטיה שלנו איתנה.
אנו פועלים למימוש רצונם של אזרחי ישראל באמצעות הקמת ממשלה חדשה בהקדם האפשרי - ממשלת חירום לאומית.
בכוונתי להביא לישראל מנהיגות חדשה - כזו שתפעל למען כלל אזרחי ישראל. יחד עם שותפיי, נבנה ישראל חזקה יותר: ישראל יהודית ודמוקרטית, ליברלית ושוויונית, אשר בטוחה בעתידה.
הקשר בין ישראל לארצות הברית הוא הרבה מעבר לברית בין ממשלות. זוהי שותפות בין עמים, שנבנתה לאורך דורות ונשענת על ערכים דמוקרטיים משותפים. זאת, מתוך האמונה כי האומות שלנו בטוחות יותר ומשגשגות יותר כאשר הן עומדות ופועלות יחד.
תחת הנהגתי, נחדש ונעמיק את השותפות הזו, נחזק את היסודות הדו-מפלגתיים שעליהם היא נשענת זה שנים, ונעבוד לקידום המזרח תיכון כולו. נפעל לא רק למען צמצומם של סכסוכים, אלא גם למינופן של הזדמנויות ושיתופי פעולה.
ישראל תמשיך להיות שותפה אסטרטגית מרכזית ומהימנה של ארצות הברית במזרח התיכון.
נתחזק שותפות המיטיבה עם שתי האומות, בכל תחומי החיים.
אני מצפה לעבוד בשיתוף פעולה הדוק ופורה עם הנשיא טראמפ וממשלו. אנו מעריכים עמוקות את ידידותם ומחויבותם העמוקה לישראל.
יחד נוכל לקדם את האינטרסים הביטחוניים המשותפים לנו, לחזק את היציבות האזורית ולפתוח דלתות חדשות לשגשוג עבור המזרח התיכון.
זוהי ישראל שאני מאמין בה וזהו העתיד שאנו מתכוונים לבנות.
4
13
155
4,912
SIMD Crawford 🟣 retweeted
The dogma of the last decade in computer vision is that all classical algorithms should be replaced by neural networks. This has been very productive but, counterintuitively, with the latest frontier models, I find myself sometimes turning back to classical methods.
If you have the training data, the dogma still holds. But there are situations where data is hard to get, there exist good classical algorithms for solving a problem, optimization methods are just better, or where some simple code just does the thing you want.
When writing code was time consuming and error prone, training everything made sense. Now, that a frontier model can code up and test solutions quickly, getting a solution through coding can be faster than gathering training data and training a model. Consequently, I think there will be more hybrid approaches that blend trained models and code.
We now have generic, high-level, planners that can execute sequences of lower-level tasks to solve large problems. These lower-level tasks may be solved by a neural network, a classical method, or specialized code written on the fly.
It never pays to be religious about technology. The future is probably hybrid. Pick the thing that works for you.
If you have the training data, the dogma still holds. But there are situations where data is hard to get, there exist good classical algorithms for solving a problem, optimization methods are just better, or where some simple code just does the thing you want.
When writing code was time consuming and error prone, training everything made sense. Now, that a frontier model can code up and test solutions quickly, getting a solution through coding can be faster than gathering training data and training a model. Consequently, I think there will be more hybrid approaches that blend trained models and code.
We now have generic, high-level, planners that can execute sequences of lower-level tasks to solve large problems. These lower-level tasks may be solved by a neural network, a classical method, or specialized code written on the fly.
It never pays to be religious about technology. The future is probably hybrid. Pick the thing that works for you.
55
156
1,250
80,727
SIMD Crawford 🟣 retweeted
This is a Zebrafish
It has the same amount of Synapses as Qwen 3.8 27B but lacks the ability to one-shot Minecraft in Three.js
Disappointing
It has the same amount of Synapses as Qwen 3.8 27B but lacks the ability to one-shot Minecraft in Three.js
Disappointing
144
537
9,310
161,753
SIMD Crawford 🟣 retweeted
Mansour Abbas: "We support a two-state solution and complete disarmament of Hamas."
31
205
1,118
41,103
Come work with me at the Muse VR team. We’re hiring ML and RL people.
Superintelligence hits different when it’s on your face. 😎
DMs open
Superintelligence hits different when it’s on your face. 😎
DMs open
My team is hiring ML Engineers / Scientists.
Strong applicants have excelled in at least two disciplines in the past, low level and ML related, very good knowledge of current ML trends and classic ML/CV; and have 8+ years of experience (PhD may count for some).
DMs open.
0
3
49
3,119
SIMD Crawford 🟣 retweeted
My team is hiring ML Engineers / Scientists.
Strong applicants have excelled in at least two disciplines in the past, low level and ML related, very good knowledge of current ML trends and classic ML/CV; and have 8+ years of experience (PhD may count for some).
DMs open.
Strong applicants have excelled in at least two disciplines in the past, low level and ML related, very good knowledge of current ML trends and classic ML/CV; and have 8+ years of experience (PhD may count for some).
DMs open.
2
9
81
8,598
SIMD Crawford 🟣 retweeted
Texas' new attorney general Nathan Johnson has revealed that he composed the score for 67 episodes of ‘DRAGON BALL Z’ as well as 5 of the Dragon Ball movies.
(Source: variety.com/2026/artisans/…)
(Source: variety.com/2026/artisans/…)
670
15,238
163,604
3,716,280
SIMD Crawford 🟣 retweeted
Republican Senator Curtis: "Unfortunately, by putting aside our core values, we’ve let down a lot of those same Americans who recently joined our party. ... Instead of peace through strength, they saw the leader of the free world side with Putin over a strategic ally."
After the Biden era, Americans wanted an economy that worked for them. Instead, they received a hastily executed war, headline after headline of scandals, a ballooning $40 trillion debt, and unpredictable tariffs.
It’s time for Republicans to return to the principled t once defined us—and that we've steadily abandoned over the last decade.
8
114
402
29,249
PC games can withstand a glut (have an audience; require marketing $), but not mobile apps.
First, review hurdles will increase to regulate the influx.
Then, incumbents will dwindle due to hurdles.
Then killer apps will cease to exist, and finally AI&GenUI will replace them all.
First, review hurdles will increase to regulate the influx.
Then, incumbents will dwindle due to hurdles.
Then killer apps will cease to exist, and finally AI&GenUI will replace them all.
Seems that Steam is drowning in game submissions. Petafab has been 12 days in the review queue. Steam says it should take 3-5 days. Heard from mobile devs that Apple review queue is over two weeks currently. Usually just couple of days.
0
0
3
633
SIMD Crawford 🟣 retweeted
Check out the #CVPR2026 compute report.
Full report and press release: cvpr.thecvf.com/Conferences/20…
Full report and press release: cvpr.thecvf.com/Conferences/20…
7
28
153
18,390