Matthew Berman

15.4K posts

Matthew Berman banner
Matthew Berman

Matthew Berman

@MatthewBerman

The future is bright.

California เข้าร่วม Haziran 2007
1.3K กำลังติดตาม122.2K ผู้ติดตาม
ทวีตที่ปักหมุด
Matthew Berman
Matthew Berman@MatthewBerman·
This situation won’t end well for the AI industry…
English
45
15
226
41.5K
lalo 🐧
lalo 🐧@lalopenguin·
things are lighting up!
lalo 🐧 tweet media
English
1
0
5
861
Lyra Intheflesh
Lyra Intheflesh@LyraInTheFlesh·
@MatthewBerman Yep. Your coverage of this sounds like you just repeated all the talking points that are being carefully repeated. Glad to see you note this one... We need to know more. Something is really fishy here...
English
1
0
6
526
Amphora
Amphora@Am4ora·
@MatthewBerman Everyone has already forgotten it was "available until June 22." Since when does a frontier model have an experation date?
English
1
0
3
490
Matthew Berman
Matthew Berman@MatthewBerman·
Fable’s stated risk is jailbreaking it into revealing information it already knows, which are existing cyber vulnerabilities. No chance that a jailbreak could transform Fable from a passive model into an agent that’s suddenly willing and able to conduct active cyber operations or autonomous research.
English
36
5
196
19.8K
ericosiu
ericosiu@ericosiu·
Fable 5 is offline so you’re better off spending that free time learning how to get 10x more from AI from @mvanhorn. He’ll give you the following superpowers: /last30days as a super research engine The ability to make a CLI out of anything to run things faster And much more:
ericosiu@ericosiu

I spent an hour picking @mvanhorn's brain about loops, agents, and AI workflows. Here are the 5 biggest takeaways from our conversation: --- 1. A loop is just a cron job plus an LLM. We've had automation forever. A cron job runs on a schedule and follows instructions: every morning at 8 AM send the report, every night at midnight back up the database. What's new is the software can make decisions now. Instead of sending the same report every day, it picks which report matters. Instead of reviewing every pull request, it flags the ones that need a human. You no longer have to just "schedule things". You can now implement judgment directly into these tasks. --- 2. Taste matters more than coding Matt isn't a software engineer. Yet he's been shipping insanely useful products because the cost of building has collapsed. His challenge was simple: "You say you're an idea person? Prove it. You now can." For years, having an idea wasn't enough. You needed engineers to test it. Now the biggest bottleneck is just deciding what deserves to be built in the first place. The builders still win. But now the builders can be anyone. Including you ;) --- 3. Skillify everything. This was probably my favorite framework from the conversation. If you do something more than once, turn it into a skill. Notion template? Skill. Research process? Skill. Sales prep workflow? Skill. If you're doing the same thing over and over, you're usually looking at a future skill. The goal is capturing judgment so it can be reused, refined, and eventually delegated to an agent. --- 4. Agents move faster through a CLI than a webpage. Humans use interfaces because we need the visual cues. We click buttons, we scan pages, that's how our brains work. An agent doesn't need any of that. Every click is another step, every page load another chance to fail. That's why Matt built Printing Press. Instead of teaching an agent to navigate a website, you give it a command line that talks straight to the underlying API. Less friction and fewer ways for it to break. Genius, if you ask me. --- 5. Apply new ideas to your own business One workflow Matt shared stood out. He'll take a YouTube video, an article, transcripts from his sales calls, and notes from previous conversations, then have AI apply the lessons directly to his own situation. Not summarize the content. Apply it. The people who are getting the most out of AI right now are getting good at spotting their own bottlenecks and pointing these increasingly capable systems at them. --- Final thoughts The tools will keep getting better on their own. Taste won't. That part's on you, and trust me, it's worth the investment. The good news: if you're reading this, you're already ahead :) P.S. List of helpful tools Matt recommended in the comments below. Happy building!

English
4
2
52
20.5K
Kilo
Kilo@kilocode·
We tested this exact question before the suspension. Claude Fable 5 was better at planning, but once the plan was fixed, GPT-5.5 produced the same service, passed the same acceptance checks, and did it at significantly lower cost. The data might be useful context for this discussion: x.com/kilocode/statu…
Kilo@kilocode

x.com/i/article/2065…

English
2
0
11
713
atlas
atlas@creatine_cycle·
there was an article written by @edels0n that mentioned that TBPN was able to monetize heavily by putting ads on clips. this doesn't make sense and here's why : - completion rate of video on X (including short clips) is abysmally low - the percentage of people even seeing the endcard is tiny - an impression !== a view. an impression is an API call. the actual content doesn't even have to render afaik - X isn't where people go to watch video (youtube is where they go to watch) TBPN monetizes with brand awareness/association and with a high quality niche audience
atlas@creatine_cycle

no one asked and i delivered. here are my thoughts on the current state of tech media while i take a bubble bath and risk destroying my fully specced out macbook forever. i also read some fan mail. 00:00 - “New Media” should be “Corporate Media” 01:45 - The different types of Tech Media 04:05 - @foundersfund Mafia reaction 06:28 - @DylanAbruscato reaction 11:32 - Edge > format 15:25 - Clip driven growth 20:05 - @Theo reaction 23:00 - The audience for tech media is small 26:24 - @jasminewsun reaction and personal style > polish 28:10 - @ti_morse reaction 31:35 - MOTS fan mail 32:08 - @0xluffy reaction 33:29 - @squirtle_says reaction 35:10 - @hopes_revenge reaction 36:12 - @DeepDishEnjoyer shoutout

English
4
1
14
1.9K
Kapisch
Kapisch@onekapisch·
@MatthewBerman But how can we even get closer to a model like Opus 4.8 locally?
English
4
0
8
1.2K
Matthew Berman
Matthew Berman@MatthewBerman·
This entire situation should wake you up to how important open source models are Get your own model running locally, go into debt if you have to
English
91
30
551
31.8K