1
00:00:00,000 --> 00:00:01,880
 Welcome to Crash News AI.

2
00:00:01,880 --> 00:00:04,300
 Today is May 8, 2025.

3
00:00:04,300 --> 00:00:06,920
 OK, forget the usual tech headlines for just a minute.

4
00:00:06,920 --> 00:00:10,640
 Seriously, what if I told you an AI just, well, basically

5
00:00:10,640 --> 00:00:14,480
 mastered web app creation better than any human benchmark

6
00:00:14,480 --> 00:00:15,400
 we've got?

7
00:00:15,400 --> 00:00:17,680
 And another one is fundamentally changing

8
00:00:17,680 --> 00:00:18,880
 how we write code itself.

9
00:00:18,880 --> 00:00:20,760
 We've got the details on these breakthroughs.

10
00:00:20,760 --> 00:00:23,400
 And yeah, they will impact how you build, test, maybe even

11
00:00:23,400 --> 00:00:24,880
 how you think about software.

12
00:00:24,880 --> 00:00:27,080
 But before we dive into all that game-changing stuff,

13
00:00:27,080 --> 00:00:29,600
 just a quick heads up on supporting Crash News AI.

14
00:00:29,600 --> 00:00:31,600
 If you're finding these deep dives valuable,

15
00:00:31,600 --> 00:00:34,040
 the easiest way, totally free, is just to rate, like,

16
00:00:34,040 --> 00:00:36,000
 and subscribe to the show on whatever platform

17
00:00:36,000 --> 00:00:36,880
 you're listening on.

18
00:00:36,880 --> 00:00:39,160
 We'd also love for you to jump into our growing Discord

19
00:00:39,160 --> 00:00:41,640
 community, connect with other developers there.

20
00:00:41,640 --> 00:00:43,600
 And for those interested in special perks,

21
00:00:43,600 --> 00:00:45,640
 becoming a Patreon is another fantastic way

22
00:00:45,640 --> 00:00:47,800
 to help us keep the lights on.

23
00:00:47,800 --> 00:00:49,840
 All right, so today we're jumping straight

24
00:00:49,840 --> 00:00:52,460
 into the AI news that we think is most impactful for you

25
00:00:52,460 --> 00:00:53,800
 as software developers.

26
00:00:53,800 --> 00:00:56,720
 We're going to try and present it in order of significance,

27
00:00:56,720 --> 00:00:58,940
 and we'll ease into the more complex bits.

28
00:00:58,940 --> 00:01:00,400
 And yeah, if there are key terms,

29
00:01:00,400 --> 00:01:01,860
 we'll definitely pause and clarify.

30
00:01:01,860 --> 00:01:04,560
 Our mission, really, is just to cut through the noise

31
00:01:04,560 --> 00:01:07,940
 and give you the insights that actually matter to your work.

32
00:01:07,940 --> 00:01:10,880
 OK, let's get into what feels like the biggest story

33
00:01:10,880 --> 00:01:11,720
 right now.

34
00:01:11,720 --> 00:01:16,700
 Google launching this Gemini 2.5 Pro preview, the I/O edition.

35
00:01:16,700 --> 00:01:19,020
 This doesn't feel like just another small step up.

36
00:01:19,020 --> 00:01:20,440
 It seems, well, pretty major.

37
00:01:20,440 --> 00:01:21,720
 Yeah, it really does.

38
00:01:21,720 --> 00:01:23,960
 What's immediately kind of grabbing headlines

39
00:01:23,960 --> 00:01:26,360
 is how well it's performing right out of the box.

40
00:01:26,360 --> 00:01:29,900
 It apparently hit number one on the Web Dev Arena leaderboard.

41
00:01:29,900 --> 00:01:30,380
 Right.

42
00:01:30,380 --> 00:01:32,420
 And for anyone who doesn't know, that leaderboard

43
00:01:32,420 --> 00:01:34,540
 isn't just about generating code snippets, is it?

44
00:01:34,540 --> 00:01:35,740
 No, not at all.

45
00:01:35,740 --> 00:01:38,820
 It specifically measures how well an AI can build,

46
00:01:38,820 --> 00:01:41,700
 like complete web applications.

47
00:01:41,700 --> 00:01:45,040
 applications, things that actually work and look good, aesthetics and functionality.

48
00:01:45,040 --> 00:01:46,040
 Exactly.

49
00:01:46,040 --> 00:01:47,320
 And it's not just web dev, apparently.

50
00:01:47,320 --> 00:01:52,320
 The reports say it's topping other El Arena leaderboards too, coding generally, vision,

51
00:01:52,320 --> 00:01:53,480
 even creative writing.

52
00:01:53,480 --> 00:01:54,560
 That's what they're claiming.

53
00:01:54,560 --> 00:01:58,880
 It looks like Google has made some serious improvements across the board.

54
00:01:58,880 --> 00:02:02,320
 But for us developers, the coding aspect is obviously huge.

55
00:02:02,320 --> 00:02:04,120
 So okay, let's unpack that.

56
00:02:04,120 --> 00:02:10,000
 If an AI can now demonstrably outperform humans on these specific benchmarks for creating

57
00:02:10,000 --> 00:02:12,240
 web apps, where does that leave us?

58
00:02:12,240 --> 00:02:13,860
 What does that shift mean?

59
00:02:13,860 --> 00:02:16,600
 That is the million dollar question, isn't it?

60
00:02:16,600 --> 00:02:24,200
 If the AI handles the, let's call it foundational creation, maybe our value shifts more towards

61
00:02:24,200 --> 00:02:29,720
 high level architectural design or complex integrations or really deeply understanding

62
00:02:29,720 --> 00:02:34,120
 user needs and translating those into specs the AI can then build out.

63
00:02:34,120 --> 00:02:37,920
 Or maybe even evaluating and refining what the AI produces, making sure it meets those

64
00:02:37,920 --> 00:02:41,200
 really specific nuanced requirements.

65
00:02:41,200 --> 00:02:44,140
 The improvements aren't just front end or UI.

66
00:02:44,140 --> 00:02:49,680
 The sources are pointing to big gains in code transformation to refactoring huge code bases,

67
00:02:49,680 --> 00:02:55,020
 also editing existing code, and even managing these complex agentic workflows.

68
00:02:55,020 --> 00:02:56,020
 Agentic workflows.

69
00:02:56,020 --> 00:02:59,520
 That's where you have multiple AI agents kind of collaborating on a bigger coding task.

70
00:02:59,520 --> 00:03:00,520
 Exactly.

71
00:03:00,520 --> 00:03:04,200
 Moving beyond just single shot generation toward like more sophisticated problem solving.

72
00:03:04,200 --> 00:03:05,200
 Okay.

73
00:03:05,200 --> 00:03:06,200
 And it's not just code either.

74
00:03:06,200 --> 00:03:10,400
 We're looking at this state-of-the-art video understanding scored really high on some video

75
00:03:10,400 --> 00:03:11,400
 ME benchmark.

76
00:03:11,400 --> 00:03:12,400
 Yeah.

77
00:03:12,400 --> 00:03:13,400
 84.8%.

78
00:03:13,400 --> 00:03:18,600
 Which, you know, might seem unrelated at first, but that ability to deeply understand video

79
00:03:18,600 --> 00:03:20,600
 content suggests a broader intelligence.

80
00:03:20,600 --> 00:03:25,560
 If you connect the darts, great at coding, understands visual data, creative writing,

81
00:03:25,560 --> 00:03:28,260
 it hints at a really powerful underlying capability.

82
00:03:28,260 --> 00:03:33,160
 So maybe potential for AI that can understand multimodal inputs in our dev environments

83
00:03:33,160 --> 00:03:34,160
 down the line.

84
00:03:34,160 --> 00:03:35,160
 Could be.

85
00:03:35,160 --> 00:03:36,620
 And the key thing is this isn't just research.

86
00:03:36,620 --> 00:03:42,340
 It's apparently accessible now via the Gemini API, Vertex AI, AI Studio, even the Gemini

87
00:03:42,340 --> 00:03:43,540
 Chatbot app.

88
00:03:43,540 --> 00:03:44,540
 Right.

89
00:03:44,540 --> 00:03:48,340
 And releasing this just before Google I/O on May 20th, that feels very intentional.

90
00:03:48,340 --> 00:03:49,340
 Oh, definitely.

91
00:03:49,340 --> 00:03:50,340
 They're setting the stage.

92
00:03:50,340 --> 00:03:52,220
 I assume more AI announcements are coming.

93
00:03:52,220 --> 00:03:54,740
 It really feels like they're signaling a major shift.

94
00:03:54,740 --> 00:03:59,460
 But yeah, the core takeaway for me, the really shocking part, an AI model demonstrably beating

95
00:03:59,460 --> 00:04:03,260
 humans on coding and web app benchmarks is something.

96
00:04:03,260 --> 00:04:04,820
 OK, let's pivot a bit.

97
00:04:04,820 --> 00:04:06,220
 Some big news from OpenAI, too.

98
00:04:06,220 --> 00:04:09,620
 They've got a restructuring going on and a pretty hefty acquisition.

99
00:04:09,620 --> 00:04:11,220
 Yeah, the restructuring is interesting.

100
00:04:11,220 --> 00:04:16,860
 They're trying to formalize that balance between their original nonprofit mission and the realities

101
00:04:16,860 --> 00:04:19,060
 of being a huge R&D player.

102
00:04:19,060 --> 00:04:25,140
 So the nonprofit stays in control, but the for-profit arm becomes a public benefit corporation.

103
00:04:25,140 --> 00:04:28,500
 The idea seems to be aligning profit more explicitly with the mission.

104
00:04:28,500 --> 00:04:29,900
 An interesting structure.

105
00:04:29,900 --> 00:04:34,700
 And then this acquisition, Winsurf, a coding startup, for a reported $3 billion.

106
00:04:34,700 --> 00:04:35,780
 $3 billion, yeah.

107
00:04:35,780 --> 00:04:37,140
 That's a massive bet.

108
00:04:37,140 --> 00:04:43,240
 Winsurf, which you might know as Exofunction or Codenium previously, they build AI tools

109
00:04:43,240 --> 00:04:44,500
 to help write code.

110
00:04:44,500 --> 00:04:45,940
 So bringing them in-house.

111
00:04:45,940 --> 00:04:50,460
 OpenAI is basically jumping directly into the AI coding assistant market, competing with

112
00:04:50,460 --> 00:04:52,420
 tools like Cursor, Replit.

113
00:04:52,420 --> 00:04:53,420
 Looks like it.

114
00:04:53,420 --> 00:04:54,420
 It's a very clear signal.

115
00:04:54,420 --> 00:04:56,060
 They're not just providing the base models anymore.

116
00:04:56,060 --> 00:04:58,300
 They want to be in the developer tool chain directly.

117
00:04:58,300 --> 00:04:59,300
 Wow.

118
00:04:59,300 --> 00:05:00,820
 So one of the biggest names in AI...

119
00:05:00,840 --> 00:05:03,600
 by making a direct play for developer tools.

120
00:05:03,600 --> 00:05:06,720
 How could that shake things up for our day-to-day workflows?

121
00:05:06,720 --> 00:05:10,000
 - Well, it could mean much tighter integration, right?

122
00:05:10,000 --> 00:05:12,560
 Imagine having the power of something like GPP

123
00:05:12,560 --> 00:05:15,460
 deeply embedded, maybe even optimized

124
00:05:15,460 --> 00:05:17,060
 within your coding environment.

125
00:05:17,060 --> 00:05:18,920
 It could lead to smoother workflows,

126
00:05:18,920 --> 00:05:22,360
 perhaps entirely new ways of generating code, debugging,

127
00:05:22,360 --> 00:05:24,040
 maybe even collaborating.

128
00:05:24,040 --> 00:05:26,080
 It's definitely gonna change the landscape.

129
00:05:26,080 --> 00:05:27,960
 - Yeah, one to watch closely.

130
00:05:27,960 --> 00:05:29,080
 Okay, let's shift gears again

131
00:05:29,080 --> 00:05:30,960
 this time to the open source world.

132
00:05:30,960 --> 00:05:35,640
 Litrix just dropped an AI video model, LTX video 13B.

133
00:05:35,640 --> 00:05:38,500
 - Right, open source in the video generation space

134
00:05:38,500 --> 00:05:41,520
 feels like a big step towards democratizing it.

135
00:05:41,520 --> 00:05:42,340
 - Definitely.

136
00:05:42,340 --> 00:05:45,200
 And the claims are pretty bold, speed first off,

137
00:05:45,200 --> 00:05:47,560
 up to 30 times faster than similar models.

138
00:05:47,560 --> 00:05:48,440
 - That's what they're saying.

139
00:05:48,440 --> 00:05:52,440
 And crucially, it apparently runs on consumer GPUs.

140
00:05:52,440 --> 00:05:54,200
 That's huge for accessibility.

141
00:05:54,200 --> 00:05:55,680
 - Yeah, it means developers like you and me

142
00:05:55,680 --> 00:05:57,680
 could actually run it without needing

143
00:05:57,680 --> 00:05:59,040
 a massive server farm.

144
00:05:59,040 --> 00:05:59,880
 - Exactly.

145
00:05:59,880 --> 00:06:00,880
 And it's not just fast.

146
00:06:00,880 --> 00:06:02,960
 They're talking about features like multi-scale rendering

147
00:06:02,960 --> 00:06:05,560
 for better detail, plus upgraded controls

148
00:06:05,560 --> 00:06:08,280
 for camera motion, key frames, sequencing.

149
00:06:08,280 --> 00:06:09,880
 - Sounds like they're trying to give developers

150
00:06:09,880 --> 00:06:12,520
 real creative control, not just a black box.

151
00:06:12,520 --> 00:06:13,360
 - Seems that way.

152
00:06:13,360 --> 00:06:15,320
 And another important point,

153
00:06:15,320 --> 00:06:18,800
 it's trained on licensed data from Getty and Shutterstock.

154
00:06:18,800 --> 00:06:20,700
 That addresses a lot of the commercial use

155
00:06:20,700 --> 00:06:21,760
 and copyright concerns.

156
00:06:21,760 --> 00:06:22,600
 - That's smart.

157
00:06:22,600 --> 00:06:25,000
 So making it available on Hugging Face and GitHub

158
00:06:25,000 --> 00:06:27,320
 with a free tier for smaller orgs,

159
00:06:27,320 --> 00:06:29,600
 that really lowers the barrier to entry.

160
00:06:29,600 --> 00:06:31,600
 - Yeah, it opens up possibilities for developers

161
00:06:31,600 --> 00:06:35,900
 to integrate AI video features into apps and tools

162
00:06:35,900 --> 00:06:37,640
 where it just wasn't feasible before.

163
00:06:37,640 --> 00:06:39,060
 A real opportunity there.

164
00:06:39,060 --> 00:06:41,560
 - Okay, moving into a more specialized area now.

165
00:06:41,560 --> 00:06:44,760
 Now, Cognition announced something called Kevin32B.

166
00:06:44,760 --> 00:06:48,360
 It's an AI model specifically for generating CUDA kernels.

167
00:06:48,360 --> 00:06:49,360
 Right.

168
00:06:49,360 --> 00:06:54,920
 Now, for folks maybe not deep in GPU programming, CUDA kernels are basically these small, highly

169
00:06:54,920 --> 00:06:57,800
 parallel programs that run on NVIDIA GPUs.

170
00:06:57,800 --> 00:07:02,360
 They're absolutely critical for speeding up tasks in machine learning, scientific computing,

171
00:07:02,360 --> 00:07:05,240
 advanced graphics, anything really computationally heavy.

172
00:07:05,240 --> 00:07:07,720
 Okay, so super important for high performance stuff.

173
00:07:07,720 --> 00:07:09,640
 What's special about Kevin32B then?

174
00:07:09,640 --> 00:07:12,700
 The interesting bit seems to be how it's trained.

175
00:07:12,700 --> 00:07:17,640
 It uses reinforcement learning to optimize the feedback it gets during training.

176
00:07:17,640 --> 00:07:22,140
 Think of it like the AI learning through trial and error, but specifically focused on making

177
00:07:22,140 --> 00:07:27,320
 the generated CUD code faster and more correct compared to older methods.

178
00:07:27,320 --> 00:07:30,180
 Ah, so it's not just spitting out code.

179
00:07:30,180 --> 00:07:35,560
 It's learning to generate better, more optimized code for this very specific, very performance-critical

180
00:07:35,560 --> 00:07:36,560
 task.

181
00:07:36,560 --> 00:07:37,560
 Exactly.

182
00:07:37,560 --> 00:07:41,160
 And it's got a real, tangible impact on the speed of a whole lot of applications that

183
00:07:41,160 --> 00:07:43,420
 rely on GPU acceleration.

184
00:07:43,420 --> 00:07:44,420
 Makes sense.

185
00:07:44,420 --> 00:07:48,760
 Okay, let's broaden out again and look at how AI is creeping into our, well, our operating

186
00:07:48,760 --> 00:07:49,760
 systems.

187
00:07:49,760 --> 00:07:51,840
 Microsoft's got AI updates for Windows 11.

188
00:07:51,840 --> 00:07:56,920
 Yeah, this is that trend of AI becoming less of a standalone app and more just part of

189
00:07:56,920 --> 00:07:58,240
 the OS fabric.

190
00:07:58,240 --> 00:08:02,280
 The start menu update is mostly cosmetic, but the new AI features are popping up in

191
00:08:02,280 --> 00:08:03,280
 various places.

192
00:08:03,280 --> 00:08:07,080
 Right, like AI and Windows settings to help adjust things or troubleshoot.

193
00:08:07,080 --> 00:08:09,160
 To fill in paint, which is kind of wild.

194
00:08:09,160 --> 00:08:10,160
 Uh-huh.

195
00:08:10,160 --> 00:08:13,520
 And a relay feature in Photos, better text copying and screenshotting in the Snipping

196
00:08:13,520 --> 00:08:14,520
 Tool.

197
00:08:14,520 --> 00:08:19,480
 Even File Explorer gets right-click shortcuts for image editing and summarizing text.

198
00:08:19,480 --> 00:08:21,000
 It feels like it's everywhere.

199
00:08:21,400 --> 00:08:26,600
 Now, these aren't direct coding tools, obviously, but think about the little time savers, quickly

200
00:08:26,600 --> 00:08:31,280
 grabbing part of a screen with text, summarizing some documentation you just downloaded.

201
00:08:31,280 --> 00:08:32,280
 Those things add up.

202
00:08:32,280 --> 00:08:36,380
 Yeah, it could subtly streamline workflows, make things a bit smoother, even if it's not

203
00:08:36,380 --> 00:08:38,240
 writing code for you directly.

204
00:08:38,240 --> 00:08:39,240
 Exactly.

205
00:08:39,240 --> 00:08:44,760
 And note that initial rollout is for Windows Insiders on those specific Snapdragon X Copilot

206
00:08:44,760 --> 00:08:49,800
 plus PCs, suggests a link between these features and hardware capabilities too.

207
00:08:49,800 --> 00:08:50,800
 Interesting.

208
00:08:50,800 --> 00:08:55,840
 Very gradual integration, but definitely points to AI being more pervasive in our daily computing.

209
00:08:55,840 --> 00:09:00,600
 All right, let's touch on a super critical area, cybersecurity, seeing some new open

210
00:09:00,600 --> 00:09:02,240
 source AI tools pop up there too.

211
00:09:02,240 --> 00:09:06,600
 Yeah, this is a space where AI's ability to sift through massive amounts of data really

212
00:09:06,600 --> 00:09:07,720
 shines.

213
00:09:07,720 --> 00:09:12,360
 Cisco launched Foundation Sec 8B, a cybersecurity-specific model.

214
00:09:12,360 --> 00:09:17,080
 Met has expanded its AI defenders suite with real-time threat filters, and Project Discovery's

215
00:09:17,080 --> 00:09:19,400
 Nuclei tool continues to be important.

216
00:09:19,400 --> 00:09:24,120
 The main goal here is just giving security teams better tools to react faster to threats,

217
00:09:24,120 --> 00:09:25,120
 right?

218
00:09:25,120 --> 00:09:26,880
 Because the threats are getting more sophisticated too.

219
00:09:26,880 --> 00:09:27,880
 Precisely.

220
00:09:27,880 --> 00:09:30,540
 Help them detect and respond more effectively.

221
00:09:30,540 --> 00:09:32,720
 And the open source aspect is key here.

222
00:09:32,720 --> 00:09:37,520
 It promotes transparency, community collaboration, and makes powerful tools potentially more

223
00:09:37,520 --> 00:09:38,520
 accessible.

224
00:09:38,520 --> 00:09:42,640
 And while we might not use these tools directly while coding, the security landscape obviously

225
00:09:42,640 --> 00:09:45,480
 impacts how we have to build and deploy our software.

226
00:09:45,480 --> 00:09:46,480
 Absolutely.

227
00:09:46,480 --> 00:09:47,480
 It's all connected.

228
00:09:47,480 --> 00:09:48,480
 Okay.

229
00:09:48,480 --> 00:09:49,480
 We're seeing AI effectively.

230
00:09:49,480 --> 00:09:50,480
 Right.

231
00:09:50,480 --> 00:09:51,480
 Prompt engineering.

232
00:09:51,480 --> 00:09:53,040
 Seems like there's always more to learn there.

233
00:09:53,040 --> 00:09:58,120
 There was a highlighted prompt for ChatGPT focusing on intellectual sparring.

234
00:09:58,120 --> 00:09:59,120
 Yeah.

235
00:09:59,120 --> 00:10:00,120
 This is really interesting.

236
00:10:00,120 --> 00:10:01,720
 It gets at how we need to evolve our interaction with these.

237
00:10:01,720 --> 00:10:07,720
 these LLMs. This specific prompt basically tells the AI, don't just agree with me. Analyze

238
00:10:07,720 --> 00:10:12,520
 my assumptions, give me counterpoints, test my reasoning, offer alternatives, act like

239
00:10:12,520 --> 00:10:13,520
 a thinking partner.

240
00:10:13,520 --> 00:10:18,440
 So instead of just asking, how do I do X? You're saying, let's discuss X, challenge

241
00:10:18,440 --> 00:10:21,640
 my approach, getting it to engage more deeply.

242
00:10:21,640 --> 00:10:27,300
 Exactly. For developers using AI for brainstorming architecture, debugging tricky issues, even

243
00:10:27,300 --> 00:10:33,440
 understanding complex concepts, that kind of interaction can unlock way more value.

244
00:10:33,440 --> 00:10:37,520
 It's a good reminder that the output quality really depends on how well you frame the input,

245
00:10:37,520 --> 00:10:40,800
 how you prompt it. It's a skill in itself.

246
00:10:40,800 --> 00:10:41,960
 Definitely an essential skill now.

247
00:10:41,960 --> 00:10:45,120
 All right, just to round things out, let's do a quick scan of some other AI tools and

248
00:10:45,120 --> 00:10:47,960
 resources that caught our eye, just for broader awareness.

249
00:10:47,960 --> 00:10:54,600
 Sure. We're seeing progress everywhere. Suno v4.5 for AI music generation keeps improving.

250
00:10:54,600 --> 00:10:59,200
 The journey's omni-reference is trying to tackle consistent image generation from reference

251
00:10:59,200 --> 00:11:00,200
 images.

252
00:11:00,200 --> 00:11:02,120
 Hmm. Consistency is a big challenge there.

253
00:11:02,120 --> 00:11:06,920
 It is. Then you've got things like Kripal, an AI agent for video creation, Nyoma for

254
00:11:06,920 --> 00:11:12,020
 AI sales analytics, Voice Panel for analyzing product feedback using AI.

255
00:11:12,020 --> 00:11:15,400
 So tools touching lots of different business areas, which might eventually filter into

256
00:11:15,400 --> 00:11:18,280
 the software we build or the requirements we get.

257
00:11:18,280 --> 00:11:22,080
 Exactly. And some useful resources too, articles breaking down different prompt engineering

258
00:11:22,080 --> 00:11:25,600
 frameworks, explainers on competitors like Mistral AI.

259
00:11:25,600 --> 00:11:31,320
 And that piece on AI reshaping student writing, that has long-term implications for the talent

260
00:11:31,320 --> 00:11:32,320
 pipeline, right?

261
00:11:32,320 --> 00:11:36,640
 Potentially, yeah. How future developers learn and think might be influenced by these tools

262
00:11:36,640 --> 00:11:37,640
 from early on.

263
00:11:37,640 --> 00:11:41,440
 Okay. And finally, just a few super quick news hits from the wider AI world.

264
00:11:41,960 --> 00:11:48,280
 Yeah, rapid fire. NBC Sports using an AI voice of a former narrator, Netflix revamping its

265
00:11:48,280 --> 00:11:54,120
 TV interface with generative AI, the U.S. pushing for more AI chip tracking to limit smuggling,

266
00:11:54,120 --> 00:11:58,120
 the ongoing Musk versus OpenAI legal saga. Always drama there.

267
00:11:58,120 --> 00:12:02,840
 Always. And more hardware competition like Huawei aiming to compete with

268
00:12:02,840 --> 00:12:05,960
 Nvidia's high-end AI chips. Okay, wow, that's a lot.

269
00:12:11,160 --> 00:12:15,640
 It feels like today's deep dive really shows this rapid, almost startling acceleration in what AI

270
00:12:15,640 --> 00:12:20,840
 can do. Understanding code, generating code, creating media, even weaving itself into our

271
00:12:20,840 --> 00:12:26,360
 operating systems, the Gemini 2.5 Pro News, especially hitting those benchmarks, and OpenAI's

272
00:12:26,360 --> 00:12:30,680
 big moves into developer tooling, those feel particularly significant for anyone building

273
00:12:30,680 --> 00:12:36,120
 software right now. Absolutely. The pace is just incredible. And it presents this dual reality,

274
00:12:36,120 --> 00:12:41,080
 right? Huge opportunities, but also potential disruptions to how we work. Staying informed,

275
00:12:41,080 --> 00:12:45,880
 really trying to understand what these tools can and can't do yet, that's going to be crucial.

276
00:12:45,880 --> 00:12:51,160
 Yeah, absolutely critical. So here's maybe a final thought to chew on as you go about your day.

277
00:12:51,160 --> 00:12:56,120
 With AI now demonstrably hitting top performance on coding benchmarks,

278
00:12:56,120 --> 00:13:01,880
 how does that fundamentally change your role? What does being a software developer look like

279
00:13:01,880 --> 00:13:07,880
 in a few years? Is it more about augmentation? Is it deeper collaboration with these AI tools?

280
00:13:07,880 --> 00:13:11,480
 Or is it shifting towards something entirely new we haven't even defined yet?

281
00:13:11,480 --> 00:13:15,160
 Heavy questions, but important ones to start thinking about now.

282
00:13:15,160 --> 00:13:18,840
 Definitely. We really encourage you to check out some of the resources we mentioned,

283
00:13:18,840 --> 00:13:21,880
 dig deeper, and consider these questions in the context of your own work.