Spiegs (@FlippyMeister) - Perfil do Twitter | Zamantika Mersobahis Locabet

Tweet fixado

Spiegs@FlippyMeister·29 May

Find me my next gig and I’ll give you $5,000. I’m based in SF, product background, both technical and user-facing products. Great track record at companies ranging from AdTech to crypto to DTC & peptides. TLDR, I’m good at making products 10x better, boosting revenue, sign-ups, engagement, and keeping users very happy. Engineers and designers love working with me. Pretty technical, vibe coder, and built my first AI eval in harbor recently. Open to: - Product roles - Growth / Marketing - Anything interesting where there are hard problems to solve. DM me if you’re either hiring or know someone who is.

English

1

0

12

9.5K

Spiegs@FlippyMeister·1h

@ravikiran_dev7 … both tbh CLI for prompting/execution, and then I’ll look at GUI to monitor the work when it’s design-related.

English

0

23

Ray🫧@ravikiran_dev7·12h

AI is splitting into two camps: GUI users Vs CLI users Which side are you on?

English

173

11

286

65.9K

Spiegs@FlippyMeister·1h

@signulll kind of funny to me how Google was SO early on AI, Meta had whole teams of AI researchers and they couldn’t think of anything to apply it other than ads and feeds.

English

0

93

signüll@signulll·3h

meta is narratively homeless. they own the largest distribution surface in human history & yet they don't live anywhere that the ppl minting the next decade's mythology actually congregate. that's a problem for recruiting the few good ppl they still want, & for everyone betting on where the future gets built. you can't be the future from outside the layer that decides what counts as it. if i were them, i'd acquire some ppl who understand how to craft a narrative on this medium. google briefly did it but now fell flat again.

Polymarket@Polymarket

JUST IN: Meta’s CTO says morale is near “the worst it’s ever been” — leadership will offer increased snack budgets to lift spirits.

English

35

16

492

64.5K

Spiegs@FlippyMeister·14h

@vasuman It’s absolutely terrible. Much better within Droid, but you can’t use your sub unfortunately.

English

0

315

vas@vasuman·18h

They lobotomized Claude in Claude Code. Worst I've ever seen. Completely ignoring instructions on a fresh context window using Opus 4.8 xHigh. Every time I've called this out in the past they respond saying NUH UH and then publish some bs 2 weeks later saying oops teehee.

English

57

8

449

54.9K

Spiegs@FlippyMeister·14h

When it came to “normal” front-end work, Fable was cracked. Really high-quality design. Opus 4.8 is much better at *design* vs GPT, but Fable was a whole different level. It also fixed Opus 4.7-4.8’s unreliability. Sure, it would do things you didn’t ask it to do, but usually it was actually useful. In my case, it added agent instructions to another harness, which I did not ask it to do, but it was relatively harmless and immediately reversed the change after I asked. Almost a week later, I realized that it baked things into my product that didn’t even cross my mind. Thought of edge cases and screens that we should have had in certain edge cases. Sure, I didn’t have a clue that it did it until now lmao. But it did! I also had it working on a different site and gave it product images and on its own, it correctly matched the product images to different selections without me asking. Opus doesn’t even come close on the same prompts/project. I find that Opus 4.8 can be like a toddler in your codebase, but it happens to be more ambitious than GPT and often write better plans. GPT will push back and often morph your plans into their own. But it’s extremely reliable. There are many MCPs that I wouldn’t dare give Opus models access to. My advice for now: use opus for plans/specs and design / front-end , GPT for execution on anything technical.

Matt Shumer@mattshumer_

Seeing so many people saying Fable wasn’t that much better than other models. Sure, if you’re just doing normal frontend/backend work that Opus/GPT 5.5 can already do, it’s going to feel basically the same. The magic comes when you’re asking it to do things other models can’t.

English

0

122

Spiegs@FlippyMeister·15h

@droidbuilds Claude’s good at specs and design, but I think it’s unreliable. GPT is reliable but a little autistic. Claude plays with your codebase like a toddler but makes things look nice.

English

0

147

DROID@droidbuilds·1d

How accurate is this? 😭

English

449

627

12.5K

703.5K

Spiegs@FlippyMeister·17h

@RhysSullivan AGGGHH WHY WOULD U DO THIS TO ME BRO I GOT EXCITED 😭😭😭😭😭

English

0

107

Rhys@RhysSullivan·18h

ZXX

4

1

129

7.4K

Spiegs@FlippyMeister·2d

@asaio87 I’ve done some side by sides with it and Opus and design taste and execution was FAR better with Fable

English

1

0

1

473

andrei saioc@asaio87·2d

Everybody says fable was so good and solve hard problems, and strategizes, and brainstorms funny thing is, during the 3 days it was available I only saw just some nice rendering graphics that was all I saw that was impressive about it I personally had 24 hrs with it, and built the same stuff as i do with Opus and felt the same.

English

81

2

132

31.7K

Spiegs@FlippyMeister·2d

I miss it purely for how good it was at design and front-end work. Not perfect, kinda over-hyped tbh, but great model. You have to watch Opus 4.8 like it’s a toddler or it can do some really dumb stuff.

Robin Ebers · AI for Small Business@robinebers

missing Fable 5 so much not even talking about coding it was just so much smarter at ingesting data, writing files, and connecting things Opus is ok at these things (honestly, much better than GPT-5.5), but Fable was on another level while it was around, I've processed 15+ hours of video, YouTube competitors, screen frames and thumbnails, patterns and anti-patterns it was the first model that really made me feel like I can trust it with knowing what I want, how to store it reliably the more I analyze the work that this model has produced for me, the more I realize that Fable was the first model that has a better understanding of how AI works and that resulted in it creating files and prompts that are dramatically better than anything any other model would produce hope it comes back soon because it feels like we went backwards bad

English

0

1

103

Spiegs@FlippyMeister·2d

I really miss how good it was at front-end / design work, and just how reliable it was. Bridged the gap between GPT’s technical competency and Claude’s design abilities. Far more reliable/trustworthy version of Claude. Had a tendency to do things it wasn’t asked to do, most of the time they were good or pretty harmless. Opus 4.8 has that same tendency, but in an intern taking down prod kinda way. Has to be watched like a toddler, so I use it almost exclusively for front-end design work or writing specs.

English

0

52

Robin Ebers · AI for Small Business@robinebers·3d

missing Fable 5 so much not even talking about coding it was just so much smarter at ingesting data, writing files, and connecting things Opus is ok at these things (honestly, much better than GPT-5.5), but Fable was on another level while it was around, I've processed 15+ hours of video, YouTube competitors, screen frames and thumbnails, patterns and anti-patterns it was the first model that really made me feel like I can trust it with knowing what I want, how to store it reliably the more I analyze the work that this model has produced for me, the more I realize that Fable was the first model that has a better understanding of how AI works and that resulted in it creating files and prompts that are dramatically better than anything any other model would produce hope it comes back soon because it feels like we went backwards bad

English

45

7

272

18.6K

Spiegs@FlippyMeister·2d

@ItakGol Lmao nice one But fr tho idk how this would work if even foreign-national Anthropic employees aren’t allowed to use it, and I don’t see Dario letting them go anytime soon or complying with that.

English

0

124

Itamar Golan 🤓@ItakGol·2d

What the heck

English

564

401

9.3K

2M

Spiegs@FlippyMeister·2d

@mark_k @OpenAI Same goes for Claude Cowork, but yes, super annoying.

English

0

61

Mark Kretschmann@mark_k·3d

Moving a project in @OpenAI Codex from one folder to another is messier than it should be. This feels like a basic workflow, but it clearly isn’t. I also don’t love how tightly Codex projects are tied to physical folders. It’s fine at first, but gets rigid fast when you want to reorganize things or work across multiple folders.

English

33

6

246

48.8K

Spiegs@FlippyMeister·2d

Codex in Claude Code plugin is kinda underrated Whenever claude gets stuck, I bring in codex for review and it figures it out right away

English

1

0

59

Spiegs@FlippyMeister·3d

Fair enough! Curious whether it was priming by raising the bar and telling it that it’s Fable, and it knows what fable is good at or something to do with Fable’s “voice” or if it had to do with the part of the system prompt that tells it it’s really good at orchestration and long-horizon work. My guess is voice and telling it it’s Fable could’ve had something to do with it. Going to work on my own stripped-down/modified version of Opus 4.8 system prompt and see how it does

English

0

1

800

Jamieson O'Reilly@theonejvo·3d

@FlippyMeister Yeah honestly was not a serious experiment more like a meme but i did note the fable prompt did seem to land closer to Apple UI than the vanilla 4.8

English

1

0

6

15.2K

Jamieson O'Reilly@theonejvo·4d

Did I just unlock claude-fable-5-lite? 😂 Since Fable 5 got pulled (US export control order, Anthropic is contesting it), I wanted to see how much of its character lives in the system prompt vs. the model itself. I ran the leaked Fable 5 prompt on Opus 4.8 head-to-head against stock 4.8.

English

151

243

3.7K

2.7M

Spiegs@FlippyMeister·3d

@thekitze @PeteHegseth @DeptofWar @AnthropicAI link?? 👀

English

1

0

186

kitze the 🐐@thekitze·3d

@PeteHegseth @DeptofWar @AnthropicAI what about that guy who's making gta 6 with three js?? he was like 99% done

English

6

0

203

5.7K

Pete Hegseth@PeteHegseth·3d

Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move. 🇺🇸

English

2.3K

3.2K

30.1K

3.1M

Spiegs@FlippyMeister·4d

driving a tesla is like wearing allbirds

English

0

1

41

Spiegs@FlippyMeister·4d

@JohnBuschVI @antoniogm It kind of drives bad intent. There’s a reason why incentivized ads are looked down upon

English

0

17

John Busch@JohnBuschVI·4d

@antoniogm Genuine question: do you ever see an example of how paying the user part of the ad revenue *could* work?

English

2

0

4

1.6K

Antonio García Martínez (agm.eth)@antoniogm·5d

Here comes the massive fraud. AGM's First Law of Ads: Never pay the user. Violate it at your peril.

Andrew McCalip@andrewmccalip

Looks like it's going to be a battle against spammers for the next few days. This is an interesting technical problem. There is no way they get their money out, every account traffic pattern will be reviewed and checked against fraud clusters. It's really not hard to see it. I'm more focused on lightening the load for the server endpoints, which are drowning. There is probably a CloudFlare solution I'll have to implement today. In the meantime, I'm suspending all click revenue, auto-banning all suspicious accounts, implementing user caps at $5 per hour. Thanks for all the love, this has been the craziest 24 hours on twitter since Lk99. Will do my best to keep the vibes immaculate. If you have any issues whatsoever, blanket refunds. I'll get a better advertiser telemetry report baked into the portal soon.

English

23

5

320

72.6K

Spiegs@FlippyMeister·4d

holy shit opus 4.8 is such a disgustingly horrible downgrade both for technical and design work

English

0

1

93

Spiegs@FlippyMeister·4d